2026
Commure said on 20 October 2026 that Commure Autonomous Coding had achieved Epic Toolbox designation for fully autonomous coding. The announcement describes the Toolbox designation as marking vendors that follow Epic's recommended Blueprint and integration practices, which it says streamlines integration and security review and lets coders check codes inside their existing Epic system. Commure says the product recommends ICD-10, CPT and E&M codes and modifiers with a written rationale, covering the step from a clinician's note through claim creation. The announcement reports that at Ob Hospitalist Group, described as the nation's largest OB/GYN provider group, clinicians cut charge entry time by 83 percent within three months and the software coded more than 85 percent of all charges. Tanay Tandon, Commure's chief executive, is quoted saying healthcare spends more than 250 billion dollars a year on administrative work that produces no clinical value.
Why it matters: Toolbox designation is an integration and process credential, not a measure of coding accuracy. It matters because it removes a real obstacle: coders reviewing AI output in a second application is where a lot of autonomous coding pilots quietly die. The customer figures are the part worth pressing on. An 83 percent cut in charge entry time is a measure of clinician effort, and 85 percent of charges coded is a measure of how much work the software took on, but neither one tells you whether the codes were right or whether denials went up or down. A revenue cycle leader should ask for the accuracy and denial numbers beside these two before treating the result as a win, and should note that an obstetrics hospitalist group has a narrower range of cases than a general hospital.
Source reviewed: Commure ... Company source ... Read the company brief
Philips said on 14 October 2026 that it had reached 1,000 global installations of SmartSpeed Precise, its AI-based MRI image reconstruction software. The company says the software is used in at least 95 percent of compatible clinical scanning on the systems where it is installed. Its stated performance claims are that SmartSpeed Precise delivers images up to 80 percent sharper and lets scans be completed up to three times faster compared with Philips SENSE imaging, that it can save an average of 215 hours of scanning time per system per year, and that this represents capacity for roughly 624 additional MRI examinations per system each year, or about 215,000 hours and more than 624,000 additional examinations across all 1,000 systems. Ioannis Panagiotelis, business leader for MRI at Philips, is quoted saying faster is only meaningful when it comes with the image quality clinicians need.
Why it matters: The adoption figure is the useful part of this announcement. Plenty of imaging AI gets bought, installed and then left switched off because it slows the technologist down or the radiologists do not trust the output, so a vendor claiming the software runs on nearly every eligible scan is making a claim it can be checked on. The time figures need more care. The 215 hours and the 624 extra examinations are Philips arithmetic built on its own comparison against its older SENSE method, not an independent measurement, and extra scanner capacity only turns into extra examinations if a department also has the staff and the referrals to fill it. Nothing here is a clinical claim, and faster scans with sharper pictures do not by themselves mean a better diagnosis.
Source reviewed: Philips ... Company source ... Read the company brief
Abridge said on 13 October 2026 that Abridge Assistant, which works inside the records system and is reached by chat or by voice, is now available to every clinician at every health system it serves, following a release in September. The company says the assistant pulls relevant details out of the chart with the sources cited, surfaces clinical guidance, drafts documents such as referrals and handoffs for a clinician to review, personalises notes using the patient's context, and can fold a health system's own clinical pathways into its answers. Abridge says it shows its work step by step rather than returning an answer alone, and that more than 300 enterprise health systems use its products. Abridge says it works with Epic and other records systems. Matt Troup, a physician assistant and Abridge clinical strategy principal, is quoted saying the assistant knows his patient and shows the work.
Why it matters: Drafting a note from a conversation that already happened is a contained job. Answering a question about a patient, and offering guidance, is not, because a wrong answer can reach a decision. Two things in the announcement are the ones to hold onto. The citations and the visible reasoning steps are the part an informatics team can actually audit, so ask whether a clinician can click through to the chart entry behind every claim. And the clinical pathway feature means the answers will reflect whichever version of the pathway was loaded, which makes pathway governance a safety job rather than a content job. Nothing here is a clinical claim, and Abridge describes the drafts as material for a clinician to review, not to send.
Source reviewed: Abridge ... Company source ... Read the company brief
DeepScribe said on 13 October 2026 that Comprehensive Cancer Centers of Nevada had made its ambient AI part of care across 13 treatment locations in Southern Nevada, covering 59 oncology clinicians. The announcement reports over 90 percent clinician adoption, says 80 percent of all patient encounters are documented using the platform, and says 8 in 10 clinicians were using its SmartPrep pre-visit feature by the second month. It says the rollout runs on the practice's iKnowMed records system, and describes the practice as participating in more than 150 clinical trials a year and having contributed to the development of more than 100 therapies approved by the FDA. Rupesh Parikh, the practice president, is quoted saying DeepScribe has given its physicians room back. Matthew Ko, DeepScribe's chief executive and founder, is quoted saying the speed of adoption signals the company is helping solve the right problems. The announcement gives no time-saved figure.
Why it matters: This is the first brief this site holds for DeepScribe, and the useful thing about it is what it measures. Ambient documentation tools are usually sold on minutes saved, a number that is hard to verify and easy to inflate. Adoption and share of encounters documented are harder to fake, because a clinician who dislikes the tool simply stops opening it. Treat the setting as the main caveat: oncology visits in a single practice follow familiar patterns, with the same clinicians and the same record system, which is a much easier environment than a general hospital with rotating staff. Note also that the location and clinician counts come from the vendor, and the practice's own pages describe 12 treatment centres and over 600 medical professionals, so the two organisations are counting different things.
Source reviewed: DeepScribe ... Company source ... Read the company brief
EliseAI announced preferred-partner status for inbound and outbound voice AI in Nextech practices. The initial focus is ophthalmology and dermatology, with joint availability planned for the start of 2027.
Why it matters: Connecting a phone agent to the practice schedule could reduce repeated work. A partnership announcement still leaves a practical question: when will a practice be able to use the promised integration?
Source reviewed: EliseAI ... Company source ... Read the company brief
Assort Health said on 9 October 2026 that it and Nextech will codevelop a joint AI agent offering for all Nextech customers, with deeper integration into Nextech's platform. The company says the agent handles ophthalmology patient access decisions including provider subspecialty, visit history, testing requirements, insurance status, referral status, location, urgency and clinical escalation rules. It says Assort serves 40 percent of Fortune 20 healthcare enterprises, that at DuPage Ophthalmology it supports roughly 500 patient calls a week, and that the practice saw a more than 60 percent reduction in call time. It notes Nextech was named a 2026 Best in KLAS winner for ambulatory ophthalmology solutions for a second year, and quotes practice administrator Maureen O'Malley saying what matters is whether the technology works in the reality of an ophthalmology practice.
Why it matters: Specialty scheduling is where generic voice AI usually falls over, because the rules are local and the wrong slot wastes a chair and a machine. A records system vendor codeveloping the agent means the rules can live in the system rather than in a script, and that is the part worth paying for. The only hard number here is one practice at roughly 500 calls a week, so treat the 60 percent call-time reduction as that practice's result. A joint offering is also an announcement of intent, not a shipped product, so ask when Nextech customers can actually buy it.
Source reviewed: Assort Health ... Company source ... Read the company brief
Chai Discovery announced on 9 October 2026 that GSK will get access to its platform, including its protein folding and design models, to support discovery work across GSK's pipeline. The announcement says the deal followed a wet-lab evaluation in which Chai's models generated designs against GSK targets zero-shot, without training on those targets, and that GSK then tested the designs itself. Chai's wording is that GSK conducted internal wet-lab testing of the designs, finding binders to all tested targets. Christopher Austin, GSK's senior vice president of R&D technologies, is quoted saying GSK evaluated the models across a panel of diverse targets and saw designs with strong binding affinities. Joshua Meier, Chai's co-founder and chief executive, is quoted saying the evaluation let both teams assess the models against a range of therapeutic targets in the lab. No payments, milestones or number of programs were disclosed.
Why it matters: A binder is a long way from a medicine. It means a designed molecule stuck to the thing it was aimed at, which is the first hurdle of many, and nothing here says any of it will help a patient. What makes this one worth noting is who ran the test. The lab work was done inside GSK on GSK's own targets, so the result is not the AI company grading its own homework. Zero-shot matters for the same reason: the models had not been tuned on those particular targets, which is closer to how a new program actually starts. The open question is money. No payments or program counts were disclosed, so this could be a broad licence or a small pilot, and the announcement does not say which.
Source reviewed: Chai Discovery ... Company source ... Read the company brief
Suki announced on October 8, 2026 a feature called Ask, which lets a clinician ask a spoken or typed question and get back patient details such as medications, allergies, vitals, lab results and chronic conditions without searching the chart by hand. Suki says the answers use the Wolters Kluwer UpToDate Expert AI programming interface, and are grounded only in UpToDate content maintained by more than 7,600 physician authors, editors and peer reviewers. The company says early evaluations had clinician reviewers prefer answers that included patient context 83 percent of the time.
Why it matters: Pairing the chart with a reference library is a sensible move, because a correct general answer can still be the wrong answer for one patient. The 83 percent figure comes from the company's own early evaluation, so it describes what reviewers preferred rather than what happened to a patient. The buyer question underneath it is what the feature does when the chart and the reference disagree, and who is accountable for the answer a clinician acts on.
Source reviewed: Suki ... Company source ... Read the company brief
Isomorphic Labs announced on October 7, 2026 that it has joined the Virtual Biology Initiative as a founding member to help generate AI-ready biological data. The initiative is convened by Biohub, and its stated purpose is to generate and make accessible the data enabling predictive AI models of biology. The announcement says participating organisations are collectively investing funding, data, computation and new measurement technology, and describes the effort as the largest coordinated commitment to generating AI-ready biological data to date, but it does not name the other participants or give funding figures. The stated result is an open resource for the research community. The announcement makes no clinical or regulatory claims.
Why it matters: This is a company admitting the bottleneck out loud. Protein structure prediction worked because decades of structures had already been measured and shared, and most of biology has no equivalent dataset, so models have little to learn from. Paying to generate that data and then giving it away is a bet that a bigger field beats a private advantage. Treat the largest commitment to date claim carefully, since the announcement names no partners and no figures, which makes it unverifiable from this page alone.
Source reviewed: Isomorphic Labs ... Company source ... Read the company brief
AZmed said on 7 October 2026 that the radiology department at CHU de Reims has used AZtrauma for fracture detection and AZchest for chest X-rays since 2021. The company cites a study in Pediatric Radiology covering 366 children under 18, which it says was run independently of AZmed and compared several fracture detection algorithms, reporting 93.6 percent specificity for Rayvolve and, for less experienced residents, sensitivity rising from 87.0 to 91.3 percent and specificity from 88.2 to 93.2 percent with AI help. AZmed separately cites a Radiography evaluation of 195,706 musculoskeletal X-rays, reporting an area under the curve of 0.983 for AZtrauma and 0.978 for AZchest.
Why it matters: The part worth reading twice is who ran the study. A hospital evaluation published in a journal and conducted without the vendor is stronger evidence than a company data sheet, and the Reims numbers show the biggest gain going to the least experienced readers. That is a useful way to think about where this kind of tool pays for itself. It is still one hospital and 366 children, so it describes that department rather than a promise for a busy adult trauma service.
Source reviewed: AZmed ... Company source ... Read the company brief
Hologic published on October 6, 2026 the results of an online survey of 1,000 US women aged 35 to 75, run from August 14 to 31, 2026 through the Dynata opt-in research panel and weighted to Census Bureau population estimates, with a stated sampling error of plus or minus 3.5 percentage points. Hologic says 37 percent of respondents had skipped or delayed a recommended mammogram and that 40 percent of those who delayed went more than two years without screening. It reports that 36 percent did not know preventive mammograms are covered at no cost, that 20 percent were never told they need mammography, and that half do not routinely see an OB-GYN. Hologic also reports much higher skip rates among women facing practical barriers: 59 percent among those with transportation problems against 31 percent without, and 56 percent among those with food insecurity against 29 percent without. Jyoti Gupta, chief executive of the Hologic Breast and Skeletal Health Solutions business, gave the company statement. The release does not mention AI.
Why it matters: This is a vendor-funded survey of an opt-in online panel, so treat the exact percentages as directional rather than as a census, and remember Hologic sells the equipment that benefits when screening goes up. The pattern is still the useful part: the biggest reported barriers are a ride, a stable home, food, and simply never being told, which no imaging algorithm touches. For a health system the practical reading is that outreach and transport spending may move the screening number more than another detection tool will, and the no-cost coverage gap looks like the cheapest thing to fix.
Source reviewed: Hologic ... Company research ... Read the company brief
GE HealthCare announced on October 5, 2026 an agreement to acquire SOFIE Biosciences for 945 million dollars in cash, with closing expected in the first half of 2027. SOFIE runs a United States network of 15 contract manufacturing sites operating 21 cyclotrons, plus one development and manufacturing site focused on theranostics, and it makes and distributes the radiopharmaceuticals used in precision medicine imaging and therapy. GE HealthCare said SOFIE will keep operating as an independent manufacturing partner while joining its Pharmaceutical Diagnostics segment, and that it expects the deal to add to revenue growth, adjusted margin and adjusted earnings per share in the first full year of ownership. The announcement also names a pipeline asset called FAPI-74. Artificial intelligence is not mentioned in connection with the deal.
Why it matters: This is the single clearest reading of GE HealthCare's current priorities, and it points away from software. Nearly a billion dollars in cash is going into cyclotrons and delivery routes, because a PET tracer decays on the way to the scanner and no algorithm fixes that. For an investor who bought GE HealthCare as an imaging AI story, the honest takeaway is that the company is buying supply chain, and its AI work is still a research line rather than the place the capital goes.
Source reviewed: GE HealthCare ... Company source ... Read the company brief
AZmed said on 5 October 2026 that the radiology department at AZ Sint-Lucas in Belgium is adding AZmeasure alongside AZtrauma, which the hospital has used since 2022. The company reports 97.4 percent sensitivity and 96.4 percent specificity for AZtrauma across 195,706 X-rays, and a mean error of 1.8 degrees for angles and 1.1 millimetres for distances for AZmeasure, tested across 258,373 X-rays from 100 centres in 26 countries. AZmed also calls AZtrauma the most extensively validated musculoskeletal X-ray AI in the world, which is the company's own description.
Why it matters: An existing customer widening its contract is the clearest commercial signal in imaging AI, because the department has already lived with the false positives. Automated measurement is also a different job from fracture detection: it saves a radiologist time on angles and distances rather than catching something missed, so it is judged on agreement with a human, not on sensitivity. The validation counts here are the company's, and a number of scans is not the same as a number of independent studies.
Source reviewed: AZmed ... Company source ... Read the company brief
AKASA announced on October 2, 2026 what it calls the first autonomous AI platform for inpatient medical coding and clinical documentation integrity. The company says the platform completes coding in less than 90 seconds after a patient is discharged, and that its inpatient volume grew almost six times in the past year. AKASA names Cleveland Clinic and Nebraska Methodist Health System as customers, and says its customer base covers one in ten US inpatient discharges.
Why it matters: Autonomous is a strong word in revenue cycle, because the money is not in speed alone. A coding engine that runs in 90 seconds and raises denials costs more than a slower human queue. AKASA's numbers describe volume and reach, which is useful, but the figure a finance leader needs is what happened to denial and rework rates at a named site after the switch.
Source reviewed: AKASA ... Company source ... Read the company brief
Personalis told the Securities and Exchange Commission that the waiting period for the Tempus mergers under the Hart-Scott-Rodino Act expired at 11:59 p.m. Eastern time on October 2, 2026. The filing notes Tempus refiled its antitrust notice on September 2, 2026, which restarted a fresh 30 day waiting period. The filing states the closing still depends on Personalis stockholder approval at a stockholders meeting plus the other closing conditions in the merger agreement. Tempus announced the deal on July 20, 2026 at 16.25 dollars per Personalis share, which it put at 1.5 billion dollars of enterprise value after counting the Tempus stake it already held, structured as an all-stock deal with the option to pay cash for up to half the shares.
Why it matters: Antitrust clearance removes the hurdle a buyer cannot control, and it leaves a shareholder vote that usually follows the board. For a health system already buying Tempus genomic reports, the useful question is what happens to Personalis NeXT Personal pricing, contracts and lab turnaround once one company owns both. Clearance is not closing, so nothing about today's ordering changes yet, and the deal could still be voted down or fail another condition.
Source reviewed: Personalis SEC Form 8-K ... Regulatory record ... Read the company brief
Notable published an account on 1 October 2026 of Sturdy Health using automated outreach to close breast cancer screening gaps. Reporting Sturdy Health's data across two outreach efforts, it says 940 patients were contacted and 11.6 percent engaged, that 77 percent of those who engaged completed the scheduling form, that 72 percent of engaged patients booked a mammogram, that 100 percent of appointments were booked by patients themselves with no manual staff outreach, and that everyone who rated the experience rated it positively. Omar Siddiqui, director of primary care at Sturdy Health, is quoted in the piece saying the programme is preventing breast cancer and that these patients are getting screened when they would not have been.
Why it matters: Read the funnel rather than the headline rates. An 11.6 percent engagement rate on 940 patients means roughly 109 people engaged, so the 72 percent conversion is a strong rate on a small base, and the whole exercise produced a modest number of mammograms. That is still worth doing, because it cost no staff calls at all, which is the real finding here. One framing needs care: screening does not prevent breast cancer, it finds it earlier, when treatment is usually simpler and more effective. The quoted claim is the health system's own words, and the measured result is appointments booked rather than any outcome.
Source reviewed: Notable ... Company source ... Read the company brief
Heidi says its new release can prepare charts and handle follow-up tasks across connected systems. The company describes review controls and says the release is unavailable in the EU and UK.
Why it matters: A better note helps with documentation. Completed follow-up work could remove another source of delay. The useful comparison is what gets finished safely, including tasks the AI cannot complete.
Source reviewed: Heidi ... Company source ... Read the company brief
Brainomix said on 1 October 2026 that new data from the Phase III INBUILD trial highlights the potential of quantitative CT to predict disease progression and assess treatment response in progressive pulmonary fibrosis. The trial is a Boehringer Ingelheim study, and Brainomix has described an ongoing partnership with that company in pulmonary fibrosis since May 2026. The announcement states the potential and the two intended uses but does not publish any effect sizes, accuracy figures or outcome numbers, and makes no regulatory claim for the use it describes.
Why it matters: A drug trial using an imaging measurement as a way to judge whether treatment is working is a different and more valuable business than selling software to a hospital, because a sponsor will pay for a measurement it can defend to a regulator. That is why this is worth watching even without figures. The absence of figures is also the point: predicting progression from a CT scan is a research finding here, not a cleared use, and the company's FDA-cleared e-Lung product measures features on a scan rather than forecasting a disease course. Wait for the published analysis before treating this as established.
Source reviewed: Brainomix ... Company source ... Read the company brief
Abridge described inpatient documentation cues and more detailed diagnosis support. The release also adds administrator controls and model enhancements for allergy and palliative care.
Why it matters: A missing detail can create another round of documentation work later. The useful result would be a more complete record with fewer corrections, while the clinician remains responsible for the note.
Source reviewed: Abridge ... Company source ... Read the company brief
Lunit announced on September 29, 2026 that Lunit INSIGHT MMG is expanding across the Stockholm Region, covering Capio Saint Goran Hospital, Karolinska University Hospital and Sodersjukhuset, where it works as an independent second reader. The company says the expanded deployment is expected to support between 200,000 and 250,000 examinations a year, and that the product has already supported roughly 200,000 examinations over the past three years at Saint Goran, where use began in 2023. Lunit cites a clinical study of more than 55,000 women and real-world results showing a 15 percent increase in cancer detection and more than a 36 percent reduction in radiologists' reading time. The deployment is carried out with Sectra. The announcement does not state the regulatory status covering this specific deployment.
Why it matters: A national or regional screening program is the hardest customer in imaging AI, because the volume is enormous and a miss is a public failure rather than a private one. Using AI as an independent second reader is also the most defensible design, since a human still reads every study. The detection and reading time figures come from Lunit and from Swedish practice, and European screening results do not transfer automatically to a United States population with a different case mix and a different follow up system.
Source reviewed: Lunit ... Company source ... Read the company brief
Owkin and Estonia's applied research center Metrosert announced an evaluation of K Pro within secure national infrastructure. The first phase will use selected health-data groups and explore research areas including oncology.
Why it matters: The collaboration tests whether AI research can work within national data controls. It is an evaluation plan, so useful evidence will come from repeatable analyses and the research produced.
Source reviewed: Owkin ... Company source ... Read the company brief
Assort Health said on 24 September 2026 that its AI agents now serve more than 1,300 providers across family medicine, internal medicine and pediatrics. The company reports, from its own data, that it resolves 95 percent of patient access interactions that do not require a staff handoff. It names Family Practice Associates of Lexington, Annapolis Internal Medicine, Pediatric Associates of Southwest Missouri, Voyage Healthcare, MFM Health and Pediatrica Health Group as deployments, and reports customer data of a 75 percent cut in hold times at Annapolis Internal Medicine, a 68 percent rise in staff capacity and 45 staff hours saved a month at Pediatric Associates of Southwest Missouri, and patient ratings of 4.3 out of 5 at both.
Why it matters: Named practices with named numbers is the strongest form of evidence a vendor can publish short of an independent study, because a practice administrator can be called and asked. Read the resolution figure carefully though: 95 percent is of the interactions that did not need a person, which is a narrower claim than it first sounds, and the announcement does not say how many calls were handed off. A 4.3 out of 5 patient rating is also good rather than remarkable for a phone system, and it is collected by the vendor.
Source reviewed: Assort Health ... Company source ... Read the company brief
Epic reported on September 23, 2026 that its Hello World outreach tool prevented more than 3.4 million no-shows at United States health systems in the previous 12 months. Epic cites a study at one medical center where missed appointments across ten clinics cost more than 10 million dollars a year, and cites research that only one in five Americans answer calls from numbers they do not recognise. Epic says Ochsner Health saved more than 1,000 hours of call center staff time across ten months by letting patients reschedule by text, and that health systems using Branded Calling, which shows the caller name on the patient phone, are seeing answer rates close to double.
Why it matters: A prevented missed appointment is the cheapest capacity a clinic can find, which is why access teams chase this number. But it is hard to prove, because nobody can tell whether a reminded patient would have turned up anyway, and 3.4 million across the whole Epic customer base measures the size of that base as much as the strength of the tool. The firmer number is the Ochsner Health figure of 1,000 staff hours, because saved phone time is something an access team can check against its own staffing.
Source reviewed: Epic ... Company source ... Read the company brief
GE HealthCare said on September 23, 2026 that it is expanding a research collaboration with a Mass General Brigham research team at Massachusetts General Hospital. The stated aim is a generative AI tool that could pull together structured clinical data and unstructured information, including clinical notes, images and treatment plans, so radiation oncology teams can reach a patient's treatment information faster. GE HealthCare names its existing Intelligent Radiation Therapy software as the starting point and points to possible later work across MR, CT and theranostics. The announcement describes work that could produce these capabilities rather than a cleared product, and states no regulatory status for the tool being explored.
Why it matters: A radiation oncology team loses real time hunting through records, so the problem being studied is a genuine one. The announcement is careful in a way worth noticing: it says what the collaboration will explore, not what it has proven, and it does not claim any regulatory status for the tool. A buyer should read it as a signal of where GE HealthCare is spending research money, not as something to put in next year's budget.
Source reviewed: GE HealthCare ... Company source ... Read the company brief
Heidi said on 23 September 2026 that it raised 340 million dollars, made up of a 100 million dollar Series C led by Blackbird and a 240 million dollar growth investment led by General Catalyst's Customer Value Fund, with Phoenix Court, Point72 Private Investments and Headline also taking part, at a stated valuation of 900 million dollars. The company says it supports 2.8 million patient visits every week across 190 countries and 110 languages, more than 175 million visits and 67 million clinical hours in total, and that its annual recurring revenue went from 1 million to 50 million dollars in two years. It names NHS Midlands, Metro South Health, Health New Zealand and Beth Israel Lahey Health as partners, and says the money funds work beyond documentation while medical judgment stays entirely with the human practitioner.
Why it matters: The structure of the round is the part to read. Only 100 million dollars is equity in the usual sense; the larger 240 million comes from a fund that typically finances sales and marketing growth, so this is money to go and sell rather than money to go and build. For a health system weighing a multi-year documentation contract, the useful signal is the named public health partners in the United Kingdom, Australia and New Zealand, because those buyers run long procurement and do not sign lightly. The revenue jump from 1 million to 50 million dollars is impressive and also means most customers are new, so few of them have been through a renewal yet.
Source reviewed: Heidi ... Company source ... Read the company brief
The company presented Chorus as shared AI infrastructure that can move beyond documentation into actions across clinical and administrative work.
Why it matters: The interesting part is not that Chorus can write a note. Ambience says it can act on the work around the note. That could remove a lot of repetitive work, but it also makes approval, error handling, and reversibility part of the product's real value.
Source reviewed: Ambience Healthcare ... Company source ... Read the company brief
Qventus announced on September 17, 2026 that HonorHealth selected its AI Solution Factory to co-develop custom AI tools. Qventus describes HonorHealth as Arizona's largest nonprofit health system, serving five million people in the Phoenix area with nine acute-care hospitals, more than 17,000 staff and over 4,000 affiliated providers. Qventus says the relationship expanded in stages across perioperative and inpatient care operations, and that nearly 150 hospital facilities use its products.
Why it matters: An expansion by a long-running customer is a stronger signal than a new logo, because that buyer has already seen where the product holds up and where it does not. Co-development also moves work onto the health system, so the real cost includes staff time, not only license fees. The number that settles it is not how many tools get built but which hospital tasks the AI finishes without a person picking up the leftovers.
Source reviewed: Qventus ... Company source ... Read the company brief
Assort Health said on 17 September 2026 that it expanded its integration with NextGen Enterprise EHR using NextGen Healthcare's Platinum API tier, which supports read and write transactions. The company says its agents can read available appointment slots, create, cancel and reschedule appointments, map cancellation reasons to a practice's own taxonomy, and handle referrals, patient cases, clinical notes, chart alerts, diagnoses, procedures and payment workflows. It reports more than 650,000 patient interactions across those organisations, and customer results of a 5 percent lift in appointment volume, a 115 percent increase in labour capacity, up to 4.3 million dollars of revenue captured per 100 providers and 97 percent resolution rates.
Why it matters: Write access is the whole story. Most voice AI in healthcare stops at reading the schedule and leaving a task for a human, so the staff saving never arrives. Mapping cancellation reasons to the practice's own taxonomy is the small detail that shows this was built with a real front desk in mind. The dollar figure per 100 providers is a company estimate built on assumptions about filled slots, so a practice should rebuild it from its own no-show rate rather than accept it.
Source reviewed: Assort Health ... Company source ... Read the company brief
Qure.ai said on 17 September 2026 that its clinical decision support system Aira received Class IIb CE certification under the European Union Medical Device Regulation, and the company describes it as among the first in the world to hold that certification for this kind of product. The announcement does not mention any FDA status for Aira.
Why it matters: A Class IIb certification in Europe is a real step up from the risk class that covers triage software, and it tells a European health service that the product was reviewed for a decision support role. It says nothing about the United States, where Aira is not described as cleared, so an American buyer should not read this as an FDA decision. Qure.ai's chest X-ray products and Aira are governed by separate approvals, and the paperwork does not transfer between them.
Source reviewed: Qure.ai ... Company source ... Read the company brief
Epic said on September 15, 2026 that 25 health systems are taking part in the second Agent Factory Build-a-Thon, where their own staff build AI agents inside Epic instead of waiting for a released feature. Epic says the first Build-a-Thon was held in May 2026, training for early adopters begins in October 2026, and broader availability is planned for 2027. Epic describes agents built to spot a serious bowel illness in newborns called necrotizing enterocolitis, flag which surgeries are likely to be cancelled, review medication refill requests, and record critical care time. Epic says the ECU Health agent saves 20 hours of staff time each week, and names Derek DeYoung as the person leading Agent Factory development.
Why it matters: Letting a health system build its own agent moves both the work and the responsibility inside the hospital. That is appealing to a system with informatics staff and a problem Epic will never build a feature for, and risky for a system without them, because somebody local now owns how the agent behaves on a bad day. The 20 hours a week is the ECU Health count for one agent, so read it as an example and not a benchmark. The open question is governance: who signs off that a home-built agent is safe to leave running, and who notices when it starts drifting.
Source reviewed: Epic ... Company source ... Read the company brief
Luma Health said on 15 September 2026 that it expanded its Navigator agent with new outreach workflows that rebook missed appointments and cancellations into open times, contact patients overdue for preventive care, reach patients with active referrals, and connect to outside AI workflows through Model Context Protocol servers. Reporting customer data, it says CLS Health rescheduled 2.3 times as many cancelled appointments with agentic outreach as with standard text reminders, that 28 percent of patients contacted at SJ Health scheduled a pediatric checkup, and that Southwest Virginia Community Health Systems rescheduled four times as many cancelled appointments as in the same period of 2025. Luma says its customers have saved more than 130,000 hours using Navigator in 2026, with 28 percent of care gaps scheduled and 45 percent of missed visits rescheduled.
Why it matters: The 2.3 times figure is the one worth keeping, because it compares the agent against the thing most organisations already do, a text reminder, rather than against doing nothing. That is a fair test and a meaningful margin. The Southwest Virginia comparison is weaker, since it measures against the same period last year and a year brings other changes. A 28 percent care gap scheduling rate also means most contacted patients still did not book, so this is a lift on a hard problem rather than a solution to it.
Source reviewed: Luma Health ... Company source ... Read the company brief
GE HealthCare said on 15 September 2026 that it launched CareIntellect for Operations, a cloud software application that forecasts capacity pressure up to 72 hours ahead at unit, department and enterprise level and recommends actions on discharge, imaging, transfers and capacity. It runs on two GE HealthCare models, a Pressure Forecast model built from historical operations data plus live feeds from the records system, and an Estimated Day of Discharge model updated hourly. The Queen's Health Systems and Duke Health are named as the first health systems to implement it. GE HealthCare cites its own 2025 figures from existing Command Center customers of up to 20 million dollars in first-year cost savings, more than a one-day fall in average length of stay, and capacity for 19,000 more patients a year. The announcement does not mention FDA clearance.
Why it matters: Capacity is the constraint behind almost every other hospital problem, so a three-day warning is worth more than another dashboard if somebody acts on it. The honest caution is in the numbers: the savings and length-of-stay figures come from GE HealthCare's own Command Center customers rather than from this new product, which has two named evaluation sites and no published results yet. A forecast also only pays off where a hospital can actually change something in 72 hours, which usually means discharge planning and staffing, not bed count.
Source reviewed: GE HealthCare ... Company source ... Read the company brief
AZmed said on 15 September 2026 that it received its sixth FDA clearance, covering AZchest for detecting and locating pneumothorax, pleural effusion, consolidation and an increased cardiothoracic ratio on frontal chest radiographs. With lung nodule detection cleared in March 2025, the company says AZchest now covers five findings per image and the wider Rayvolve suite covers eight FDA-cleared findings across musculoskeletal and chest X-rays.
Why it matters: Four findings on one image in a single clearance is a real widening, because each one would otherwise be a separate purchase, a separate integration and a separate set of alerts. A clearance says the FDA accepted the submission for those findings, so the useful comparison between vendors is the finding list and the scan type, not the clearance count. An increased cardiothoracic ratio is also a measurement rather than a diagnosis, and a buyer should know which of the four will actually change a report.
Source reviewed: AZmed ... Company source ... Read the company brief
Lunit announced on September 10, 2026 a collaboration with 10x Genomics to combine Lunit SCOPE IO, its AI analysis of standard stained pathology slides, with 10x Genomics Xenium and Atera spatial platforms. The stated aim is to better understand how tissue appearance relates to the underlying biology of tumours, and to identify biomarkers associated with treatment response and resistance. The announcement carries an explicit disclaimer that Lunit SCOPE IO is for research use only and not for use in diagnostic procedures. Lunit describes the SCOPE platform as used in research partnerships with pharmaceutical and laboratory organisations for biomarker research and companion diagnostic development, separately from its FDA-cleared INSIGHT Breast Suite.
Why it matters: This is the quieter half of Lunit's business and the one that pays differently. Screening AI sells to hospitals at hospital prices, while biomarker work sells to pharmaceutical companies developing drugs, which is a smaller number of much larger contracts. For a research team the practical appeal is skipping a step, since a cheap stained slide read by AI may flag what would otherwise need expensive spatial sequencing. Read the research use only label literally: nothing here is for diagnosing a patient.
Source reviewed: Lunit ... Company source ... Read the company brief
Tempus said an award of up to $9.5 million will support development and prospective testing of an AI agent for heart-failure care.
Why it matters: The money is useful because it moves the agent into a prospective test instead of leaving it as a product idea. The interesting part will be how the study measures care, safety, and clinical responsibility. Until those results arrive, the award supports the program, not the outcome.
Source reviewed: Tempus Investor Relations ... Investor relations ... Read the company brief
UpDoc said on 9 September 2026 that ARPA-H selected it to lead development of an autonomous clinical AI system for managing cardiovascular disease under the ADVOCATE program, with an award of up to 9.2 million dollars over 39 months and with Microsoft, OpenAI and NVIDIA joining the initiative. The company describes its current FDA-cleared insulin management system as combining medical device functionality with conversational AI to support care within physician-prescribed treatment plans and defined clinical boundaries. UpDoc quotes ARPA-H programme manager Dr Haider Warraich calling ADVOCATE an unprecedented level of coordination across the federal health enterprise, coordinated with the FDA and other agencies, and cites a United States cardiovascular disease cost of more than 400 billion dollars a year.
Why it matters: The important detail for a buyer is the line between the two systems. The insulin product is cleared and sellable now; the cardiovascular one is a development award running 39 months, so nothing about it is a product a health system can buy this year. The award still matters, because a federal programme funding autonomous clinical action, coordinated with the FDA, is a signal about where that regulatory boundary may move. Until it does, treat an autonomous care claim the way you would treat any device claim and ask for the clearance that covers the exact action.
Source reviewed: UpDoc ... Company source ... Read the company brief
Epic said on 8 September 2026 that Memorial Hermann Health System clinicians are using Art, its AI assistant built into the record, to draft patient-friendly notes when they review lab and imaging results. Reporting Memorial Hermann's data, it says patients sent 32 percent fewer follow-up messages about test results when the clinician started from Art's draft, that Art has helped write more than 55,000 messages to patients, and that clinicians starting from the draft saved an average of nine seconds per result comment. Dr Ryan Walsh, chief medical information officer at Memorial Hermann, is quoted saying the draft helps produce more thorough follow-up notes that clinicians review, edit and personalise before sharing.
Why it matters: Nine seconds per comment sounds trivial and is the wrong number to judge this on. Across 55,000 messages it adds up, but the 32 percent fall in follow-up questions is the real saving, because an inbound patient message costs a clinician far more than nine seconds to answer. This also shows why Epic matters in this part of the directory: a function like this arriving inside the record a health system already owns is the main competitor to any point solution, and no new contract was signed to get it. The clinician still reviews and sends every message, so this is drafting rather than autonomous communication.
Source reviewed: Epic ... Company source ... Read the company brief
Brainomix said on September 8, 2026 that it and Endeavor BioMedicines presented results from the ENV-IPF-101 Phase 2a trial of taladegib at the European Respiratory Society Congress 2026. Brainomix says the analysis used Brainomix 360 e-Lung for AI-powered quantitative CT analysis, and that the trial showed significant treatment effects across several measures of lung structure and fibrosis. The announcement states no regulatory status for the results.
Why it matters: Being picked as the measuring tool inside a drug trial is a different kind of credit from a sales reference, because a sponsor filing can end up resting on that measurement. The results described here belong to the trial, not to the tool, so they show the measurement moved along with the drug rather than showing e-Lung is accurate on its own in routine care. For a hospital buyer the useful question is whether trial-grade readings need trial-grade scan settings that a busy clinic will not match.
Source reviewed: Brainomix ... Company source ... Read the company brief
Commure said on 3 September 2026 that Commure Ambient AI received CE marking as a Class I medical device under EU MDR 2017/745 and was registered with the United Kingdom Medicines and Healthcare products Regulatory Agency under the Medical Devices Regulations 2002. The company says that under MDR Rule 11 the product qualifies as Class I because it drafts clinical documentation rather than informing diagnostic or therapeutic decisions, with clinicians reviewing and signing every note. The announcement does not mention the FDA.
Why it matters: The reason this is Class I and not something higher is that the tool is not allowed to influence a clinical decision, and that is a useful line for a buyer to hold the vendor to. If a future version starts suggesting a diagnosis or a treatment, it is a different risk class and a different submission, and a health system should ask to be told when that happens. The marking covers Europe and the United Kingdom only, so an American buyer gets nothing from it.
Source reviewed: Commure ... Company source ... Read the company brief
Epic said on 2 September 2026 that Chart with Art, its native AI charting tool, reduced documentation time at OLVG, a teaching hospital in Amsterdam. Reporting the hospital's data, it says pediatricians cut note-writing time per appointment by 31 percent, and that overall clinician time in the chart per appointment fell 27 percent, from nearly 17 minutes to just over 12. Dr Felix Kreier is quoted saying he can focus on the conversation with the peace of mind that nothing will be forgotten.
Why it matters: A health system choosing an ambient documentation vendor should know that a comparable saving is available from the record system it already runs, with no extra contract, no separate audio retention policy and no new vendor to govern. That is not an argument against the specialist vendors, several of which report larger savings and work across more specialties, but it is the baseline the saving should be measured against. The figures are one Dutch teaching hospital's own, concentrated in pediatrics, and five minutes per appointment is a real but modest change rather than a transformation.
Source reviewed: Epic ... Company source ... Read the company brief
Candid Health said on 27 August 2026 that it is partnering with Flatiron Health to bring revenue cycle automation to oncology practices in Flatiron's network, with Flatiron integrating the platform into its OncoEMR records system. Candid Health says OncoEMR is used by more than 5,100 providers and more than 220 community oncology practices across the United States, and that it already does this work for some of the largest oncology organisations in the country.
Why it matters: Selling through the records system a practice already uses removes the hardest part of revenue cycle software, which is the integration. For a community oncology practice this is the difference between a project and a setting. The announcement names no practice that has switched, so the useful question is how many of those 220 actually turn it on and what their denial rate does afterwards. Oncology billing also leans on drug costs and prior authorisation, so general claim automation results do not transfer to it unchanged.
Source reviewed: Candid Health ... Company source ... Read the company brief
Aidoc published findings on August 25, 2026 from a white paper on Mercy's use of its aiOS platform, which Mercy launched in January 2025. Mercy commissioned AVIA Health to run an independent review of the deployment data. The review reports that flagging the most urgent cases reduced scan-to-result turnaround time from 19.8 hours to 1.3 hours, and that Mercy and Aidoc flagged 50,749 incidental findings over five months, of which 5,642 patients received accelerated clinical attention. It also states Mercy deployed 13 AI imaging solutions across more than 50 hospitals in four months, and that those solutions now analyze roughly 2.5 million patient cases a year.
Why it matters: Operational numbers at this scale are rare and genuinely useful, because most imaging AI evidence comes from one site and a few thousand scans. The reviewer being commissioned by the customer is not the same as independent research, so treat the figures as a well-documented customer account rather than a clinical finding. The number we would want next is what happened to those 5,642 patients, since faster attention is a process result and not yet a health result.
Source reviewed: Aidoc ... Company source ... Read the company brief
Innovaccer announced on August 20, 2026 that Impact Primary Care Network selected its care management solution and population health dashboards. The announcement describes Impact Primary Care Network as a physician-led, clinically integrated network serving Western North Carolina, founded by Impact Health in 2024. Allison Krivatch, the network's executive director, is quoted saying the mission is to give every primary care practice in the network the tools and infrastructure to deliver whole-person, coordinated care. The announcement gives no figures for affiliated practices, providers, patients or measured results.
Why it matters: Independent primary care networks are where value-based care is won or lost, because they carry financial risk without a hospital balance sheet behind them. A platform selection at a network founded in 2024 is an early customer rather than a reference deployment, so it says more about Innovaccer's sales reach than about results. With no practice count, provider count or patient count disclosed, there is no way to judge the size of the win from the announcement itself.
Source reviewed: Innovaccer ... Company source ... Read the company brief
Chai Discovery said on 20 August 2026 that it is collaborating with Bristol Myers Squibb on AI-driven therapeutic antibody discovery. The company says Bristol Myers Squibb will use its AI models and platform, including its molecular folding and design models, to support antibody candidate discovery across its portfolio, and describes its technology as predicting and reprogramming molecular interactions so scientists can design biomolecules with specific properties. No financial terms are disclosed in the announcement.
Why it matters: A collaboration with no disclosed money is usually an evaluation rather than a commitment, and that is worth knowing before reading it as validation. What it does show is that a large developer was willing to put its name on the work, which in antibody discovery is not free. The useful measure a year from now will be whether any of these five partners takes a Chai-designed antibody into development, because a tool that gets used but produces nothing filed is a line item, not a platform.
Source reviewed: Chai Discovery ... Company source ... Read the company brief
Luma Health said on 19 August 2026 that it is partnering with Physician's Computer Company to build its communications platform natively into PCC's pediatric electronic health record, reaching more than 300 independent pediatric practices and more than 1,400 pediatricians across the United States. The company says the integration supports two-way messaging between parents, caregivers and staff in more than 30 languages, automated messages tied to upcoming visits or sent ad hoc such as during a clinic closure, and one place for staff to manage all communications with patient and caregiver information visible. PCC's chief executive is quoted saying Luma stood out as a patient-focused technology partner.
Why it matters: Reaching small independent practices through their records vendor is the only economic way to sell them this kind of software, so the distribution is the story rather than the technology. Pediatrics is also the setting where messaging gets complicated fast, because the patient is a child and the person answering is a parent or another caregiver, sometimes with custody and consent rules attached. Support for more than 30 languages matters for the same reason. The announcement names no practice that is live, so the 300 figure is the reachable market rather than the installed base.
Source reviewed: Luma Health ... Company source ... Read the company brief
The FDA premarket notification record for K260898 names the device as Transpara (2.1.0), the applicant as Screenpoint Medical B.V., and the decision date as 14 August 2026, with a decision description of substantially equivalent. The record lists the product code as QDQ, the regulation name as radiology, and the type of submission as Special. A Special 510(k) is the route used for a modification to a device the applicant has already had cleared. The record states a clearance and does not state any finding about clinical performance. This is the seventh Transpara clearance in this site's own device records for that applicant, which also list K241831 from November 2024, K221347, K210404, K193229, K192287 and K181704 going back to 2018.
Why it matters: Clearance is a market permission, not a performance grade. The FDA found this version substantially equivalent to a device already on the market, which is a comparison to a predecessor rather than a judgement that the software helps radiologists. The reason to care is version control. ScreenPoint now has seven clearances across eight years for the same product family, and the published trial evidence people quote was generated on a particular version. Before signing anything, a buyer should make the contract name the cleared version being installed, because a product name alone no longer identifies what was tested or what was cleared.
Source reviewed: FDA 510(k) premarket notification database ... Regulatory record ... Read the company brief
Aidoc said on August 13, 2026 that its CARE Body CT Multi-Triage product received eligibility for the Medicare New Technology Add-on Payment under the Centers for Medicare and Medicaid Services Fiscal Year 2027 Medicare Hospital Inpatient Prospective Payment System final rule, effective October 1, 2026. The company describes the program as providing a Medicare fee for service add-on payment for eligible inpatient use, and does not state a dollar amount or percentage in the announcement. Aidoc states the product received FDA clearance in January 2026 and a Breakthrough Device Designation in September 2025. The payment eligibility is a Medicare coverage decision and is separate from the FDA clearance.
Why it matters: Reimbursement is the hurdle that has quietly killed good imaging AI, because a product that saves radiologist time still has to come out of someone's budget. An add-on payment moves part of the cost to Medicare for eligible inpatient cases, which makes the business case easier to write and harder to cancel. Two cautions: eligibility is not a blank cheque, since the payment applies only to qualifying inpatient stays, and the add-on payment program is time limited by design rather than permanent.
Source reviewed: Aidoc ... Company source ... Read the company brief
Hippocratic AI announced on August 13, 2026 what it calls Agentic Orchestrators, described as coordinated teams of AI voice agents supervised by a central intelligence that decides which agent engages each patient and when. The company says it offers more than 30 orchestrators across payer, provider and life sciences uses, including readmission reduction, STAR ratings improvement, chronic care management and clinical trial enrollment. Hippocratic AI states it has completed more than 250 million patient voice interactions with zero incidents of serious harm, and that its latest agent versions reached 99.89 percent correct advice with zero instances of severe harm when validated by more than 7,700 United States licensed clinicians across 775,000 calls. The company says its agents do not diagnose or prescribe, and names no customers in the announcement. No FDA status is claimed.
Why it matters: Moving from tasks to outcomes changes what a contract can promise, and a vendor willing to be paid on readmissions is making a much bigger commitment than one paid per call. The safety figures deserve care: 99.89 percent correct advice is a company-run evaluation by clinician reviewers, not a clinical trial, and 0.11 percent of 775,000 calls is still hundreds of calls judged incorrect. The absence of named customers in an announcement this specific is the gap we would close first.
Source reviewed: Hippocratic AI ... Company source ... Read the company brief
Innovaccer announced on August 13, 2026 that Mastek has joined its Platinum global systems integrator program, pairing the Innovaccer healthcare AI platform with Mastek's delivery presence across the United States, the United Kingdom, Europe and the Middle East. The announcement describes Mastek as a global provider of enterprise AI, digital and cloud services, founded in 1982 and listed on India's National Stock Exchange, operating in more than 40 countries with roughly 5,000 employees, serving more than 275 healthcare and life sciences customers at a 96 percent client retention rate. Named capability areas include revenue cycle work such as prior authorization, coding and denials, population health and value-based care, payer performance including Stars ratings, risk adjustment and utilization management, and patient access with AI agents for scheduling and referrals. No pricing, deal value or customer targets were disclosed.
Why it matters: Partnerships like this are usually read as marketing, but the useful signal is about implementation capacity rather than technology. Health systems do not fail to get value from a population health platform because the software is wrong, they fail because nobody has the staff to configure it, so adding a consultancy with thousands of people is a real answer to a real problem. The risk for a buyer is accountability: when the platform and the installer are different companies, a disappointing result can become an argument between vendors.
Source reviewed: Innovaccer ... Company source ... Read the company brief
Cadence announced on August 13, 2026 that its HypertensionOS software was selected as the second participant in the FDA pilot called Technology-Enabled Meaningful Patient Outcomes, or TEMPO. Cadence describes HypertensionOS as prescription software as a medical device with AI-assisted features that support clinician-supervised medication management. The announcement says that taking part in the pilot does not constitute FDA approval or clearance of HypertensionOS, or an FDA determination about its safety or effectiveness.
Why it matters: A pilot seat gives Cadence a direct line to the FDA while real-world data is collected, which is worth tracking if you follow which remote care tools regulators are watching. It does not say the product has been cleared, and Cadence says so itself. For a health system the useful question is what data the pilot will produce and when, because that data, not the pilot seat, is what a review committee can act on.
Source reviewed: Cadence ... Regulatory record ... Read the company brief
Assort Health said on 12 August 2026 that it launched Referrals, an AI agent for specialty practices and health systems that processes an incoming referral, checks patient eligibility, matches the patient to a provider, then contacts the patient by voice, text or email to book the appointment directly in the records system and keeps the referring provider informed. The company says it carries the referral context into later patient conversations, and that in 50 percent of referrals it identifies a better-fit provider than the one originally named. It names the Center for Sports Medicine and Orthopaedics, which it says receives roughly 75 referrals a day, reports a 77 percent conversion rate and about 90 percent of referrals automated there, and says referrals can be processed within the hour they arrive. Orthopaedics, ear nose and throat, and gastroenterology are the named specialties.
Why it matters: Speed is the whole game in referrals, because the patient who is called the same day books and the patient called next week goes somewhere else, so processing within the hour is the claim to test. The better-fit provider figure is the one to push on: an AI agent rerouting half of all referrals away from the named doctor is a clinical and a political decision inside a practice, not just an efficiency gain, and somebody has to own it. A 77 percent conversion rate at one named practice is a strong, checkable number, and it is one practice.
Source reviewed: Assort Health ... Company source ... Read the company brief
Aidoc announced on August 11, 2026 a Diagnostic AI Consortium with twelve US health systems: Advocate Health, Cedars-Sinai Health System, Hartford Healthcare, Houston Methodist, Mercy, Mount Sinai Health System, Northwell Health, Northwestern Medicine, Sutter Health, University of Florida Health, University Hospitals of Cleveland and WellSpan Health. Aidoc says the members will co-design AI-assisted diagnostic workflows, measure the effect on safety, quality and speed to diagnosis, and write shared rules for how the tools are governed. The company says its products run in nearly 2,000 hospitals, analyze 60 million patient cases a year, and that first results are expected in 2027.
Why it matters: Imaging AI is usually sold with one hospital's numbers attached, so a twelve-system plan to measure it is unusual and worth following. The value depends on whether the members publish results that reflect badly on the tools as well as results that flatter them. Until 2027 there is nothing to read, so today this is a promise to measure rather than a finding.
Source reviewed: Aidoc ... Company source ... Read the company brief
The company reported higher malnutrition code capture and expanded the automation to pressure-injury work.
Why it matters: A 57 percent increase sounds large, but the business value depends on the starting rate, the number of cases, clinical confirmation, and what happens after an audit or denial. The larger story is that one automation claim touches both care quality and revenue. That makes the measurement more important, not less.
Source reviewed: Qventus ... Company source ... Read the company brief
Recursion reported that Genentech advanced the collaboration’s first neuroscience target into a joint early discovery program alongside its quarterly business update.
Why it matters: A partner choosing to advance a target says more than another AI platform announcement. Genentech is committing joint discovery work to something found in the collaboration. It is still early, but it is the kind of step investors can track from computational discovery toward a possible drug program.
Source reviewed: Recursion Investor Relations ... Investor relations ... Read the company brief
The company announced voice agents intended to contact vulnerable members during heat, smoke, cold, and other weather emergencies.
Why it matters: This is where Hippocratic AI's safety position meets a real-world stress test. Fast outreach can help care teams reach vulnerable people before a weather event becomes a health emergency. The value depends on who gets contacted, when a human takes over, and what happens when contact fails.
Source reviewed: Hippocratic AI ... Company source ... Read the company brief
The investor hub published the company report, presentation, and analyst materials for its third quarter of fiscal 2026.
Why it matters: AI product announcements can look important without showing up in the business. Quarterly filings give investors a place to look for orders, revenue, margins, or strategy that connect those products to company performance. Siemens Healthineers is diversified, so the AI story has to be separated from everything else it sells.
Source reviewed: Siemens Healthineers Investor Relations ... Investor relations ... Read the company brief
Viz.ai published a ten year milestone announcement on July 28, 2026 stating that it operates in 2,000 hospitals, covers 230 million patients and holds 13 FDA clearances. The company says it received the first FDA De Novo authorizations for AI computer aided triage software and for ECG-based cardiovascular management software, and that it was the first company to receive Medicare reimbursement for AI software through the New Technology Add-on Payment program. All of these figures and firsts are the company's own account. The announcement does not break the 13 clearances down into individual submissions.
Why it matters: For time-critical conditions, installed base is close to the whole product, because an algorithm that is not running at the hospital a stroke patient reaches does nothing. Two notes on the wording: a De Novo is granted or authorized rather than cleared, and the two terms are not interchangeable even when a company uses them loosely. The reimbursement claim is the more durable one, because being first through Medicare's add-on payment program left behind a route other imaging AI companies now follow.
Source reviewed: Viz.ai ... Company source ... Read the company brief
PathAI announced the AISight Dx version 2.21 release on July 28, 2026. The release adds a customizable Case Assignments Dashboard with a My Cases by AI Impression widget that shows live bar charts for case prioritization, plus slide viewer changes including bulk annotation deletion, editable linear measurements and a motion-based rotation mode. It also adds faster colleague invitations with online status indicators, bulk ingestion supporting Specimen, Block ID and Slide ID fields, clinician preference management in a new Library tab, and archiving to Glacier Deep Archive with on-demand restoration. The announcement states that AISight Dx is FDA-cleared for primary diagnosis in the United States and CE-IVD marked for primary diagnosis in the European Economic Area, the United Kingdom and Switzerland.
Why it matters: Release notes rarely matter, and then one line does. Deep archive storage with restore on demand is the quiet admission that whole slide images break lab budgets, and a lab planning full digital sign out should price that before it signs. The dashboard sorting cases by AI impression is more interesting than it looks, because the order a pathologist works in starts being set by the software, and nobody has settled who is accountable when that order is wrong.
Source reviewed: PathAI ... Company source ... Read the company brief
Nabla announced Nabla Dictation for Mac on July 28, 2026, describing it as a clinical dictation application built natively for macOS that runs its full speech pipeline on the device with no audio sent to servers. The company says 44 percent of the clinicians using Nabla today work on Apple devices, and that the app works with existing dictation hardware including PowerMic and SpeechMike microphones. Nabla emphasizes a deep Epic integration and says M Health Fairview has deployed the product at scale. Chief product officer Laurent Landowski credits Apple's Core ML and Neural Engine for making the on-device pipeline possible. The announcement states no accuracy figures and makes no FDA claim.
Why it matters: Where the audio goes is the question compliance actually asks, and on-device processing answers it in one sentence instead of a data processing agreement. The honest gap is accuracy: Nabla published no error rate for this product, and on-device models are usually smaller than server models, so a careful buyer tests dictation accuracy against whatever it is replacing rather than assuming the privacy win is free. The Epic integration is the part that decides whether clinicians actually use it.
Source reviewed: Nabla ... Company source ... Read the company brief
Candid Health said on 22 July 2026 that it raised 120 million dollars in a Series D round led by Sixth Street Growth, with Oak HC/FT, 8VC and Y Combinator taking part, and that the round valued the company at three times its February 2025 Series C. The company reports 190 percent year-over-year annual contracted run-rate revenue growth and 180 percent year-over-year net dollar retention in 2025, says it serves more than 200 healthcare organisations, and frames the opportunity as the 280 billion dollars spent each year on revenue cycle management in the United States.
Why it matters: Net dollar retention of 180 percent means existing customers are spending far more each year, which is the strongest signal in this list because it is harder to manufacture than new sales. It also means the business model depends on expanding inside accounts, so a buyer should expect upsell pressure and should fix scope and pricing for more than one year. The 280 billion dollar figure is a market size, not a saving, and nothing here says what a single practice collects.
Source reviewed: Candid Health ... Company source ... Read the company brief
Luma Health said on 21 July 2026 that it launched Patient Pipeline, which reads real-time schedules, builds Google Ads campaigns aimed at open appointment slots, and traces each booking back to the campaign that produced it, while showing availability, scheduling bottlenecks and revenue potential by provider and service line. Reporting customer data, it says the University of Arkansas for Medical Sciences saw a 2.45 times return on advertising spend with one orthopaedic campaign producing 27,400 dollars in four months, that University Hospitals saw an 80 percent shorter time to service in dermatology, and that Charlotte Ear Eye Nose and Throat Associates saw an eight times return on advertising spend and 8,000 impressions from a single campaign.
Why it matters: Closing the loop from advert to booked appointment is the measurement health system marketers have never had, and it changes the argument about marketing budget from reach to revenue. Read the spread in the returns though: 2.45 times at one system and eight times at another is a four-fold difference, which says the result depends far more on the specialty and the local market than on the software. The 27,400 dollar figure over four months is also small in hospital terms, so this is a tool for filling specific gaps rather than a growth engine. Advertising open slots also raises a question of its own about which patients see the advert.
Source reviewed: Luma Health ... Company source ... Read the company brief
Nabla announced on July 21, 2026 that Brian Manning is its new chief executive officer. The company says Manning was previously president and chief revenue officer of Bamboo Health, where he helped scale the company past 140 million dollars in annual recurring revenue, and that he held early roles at PatientPing and helped build Zocdoc's enterprise sales organization. Co-founder and former chief executive Alex LeBrun becomes executive chairman and chief AI officer, continuing to lead AI strategy and product vision, while the other co-founders stay in their roles. Nabla states it supports more than 190 healthcare organizations, 100,000 clinicians and 40 million patient encounters a year, naming CVS Health, M Health Fairview, UCLA Health and University of Iowa Health Care among them. The announcement also says AMI Labs, a frontier AI research company, raised more than 1.2 billion dollars and has an exclusive partnership giving Nabla early access to its research.
Why it matters: Handing the chief executive job to a commercial leader while the founder keeps the AI work is a company telling you it has stopped worrying about whether the product works and started worrying about distribution. For a health system, that usually means better contracting and support, and it can mean sharper pricing. For a candidate, it means the hiring is likely to tilt toward sales, implementation and customer success rather than research.
Source reviewed: Nabla ... Company source ... Read the company brief
Assort Health said on 21 July 2026 that Jon Corn joined as chief revenue officer from Samsara, Kristina Kemp as vice president of customer success from Luma Health and Emil Yeargin as senior vice president of people from Gusto. The company states total funding of 222 million dollars, including a 120 million dollar Series C led by Menlo Ventures, and says its revenue grew 20 times over the previous 15 months. It reports its platform is built on more than 190 million patient interactions, 62,000 care protocols and 1.6 million decision pathways, and cites customer results of a 5 percent lift in appointment volume, a 115 percent increase in labour capacity, up to 4.3 million dollars captured per 100 providers, 97 percent resolution rates and 79 percent referral-to-appointment conversion.
Why it matters: Hiring a chief revenue officer and a head of people before anything else is the pattern of a company building a sales engine, so expect an attentive sales process and competitive pricing now, and expect upsell pressure later. The customer success hire coming from Luma Health, a competitor in the same group on this site, is the detail worth noting, because it suggests Assort is deliberately buying knowledge of how these deployments go wrong. Twenty times revenue growth off a small base is a sales number, not a performance number.
Source reviewed: Assort Health ... Company source ... Read the company brief
Chai Discovery said on 15 July 2026 that it entered a collaboration agreement with argenx, giving argenx early access to its platform for designing antibodies from scratch against therapeutic targets. The company describes Chai-2 as the first zero-shot antibody design platform to reach double-digit experimental hit rates, and says its newer Chai-3 model improves on therapeutic binding, hard-to-drug targets, multispecific design, developability and generalisation. No financial terms are stated.
Why it matters: A double-digit hit rate for designing an antibody with no prior examples would be a genuine change, because traditional discovery screens enormous libraries to find a few that bind. The figure is the company's own and the word to watch is experimental: it means something was made and tested, which is a higher bar than a computational score, but it is still Chai's own bench rather than an independent one. argenx is an antibody company with its own strong discovery engine, so it choosing to evaluate this is a more informative signal than a general technology partnership.
Source reviewed: Chai Discovery ... Company source ... Read the company brief
Chai Discovery said on 14 July 2026 that it raised 400 million dollars at a stated valuation of 3.8 billion dollars, in a round led by Index Ventures and Kleiner Perkins alongside Sequoia Capital, Dimension and other investors. The company says the money will expand the compute, data and research behind its models, speed up product development and put the technology in front of more scientists, and that its latest model, Chai-3, is being adopted across Eli Lilly, Pfizer and Novartis.
Why it matters: Spending a round this size on compute and data rather than on a laboratory is the clearest statement of what Chai Discovery is: a model company selling to drug companies, not a drug company. For a research team that is good news on access and bad news on alignment, because the vendor's incentive is adoption across many partners rather than one programme succeeding. A 3.8 billion dollar valuation on no approved product also means the next round depends on partners renewing, so ask about contract length before building a workflow on it.
Source reviewed: Chai Discovery ... Company source ... Read the company brief
Chai Discovery announced on 13 July 2026 that Novartis will get access to its AI models, including Chai-3, to support therapeutic antibody discovery across multiple targets. Financial terms were not disclosed, and the announcement gives no number of programs or milestones. Fiona Marshall, Novartis's president of biomedical research, is quoted saying AI is already reshaping how the company discovers medicines and that applying advanced models to problems like antibody design lets it explore a broader range. Joshua Meier, Chai's co-founder and chief executive, is quoted calling it a significant step forward. The announcement also names Pfizer and Eli Lilly as earlier users of Chai's models.
Why it matters: This is an access agreement, which is the lightest kind of deal in this field. A large drug company can hand a model to a research group, try it on a few targets and walk away without ever announcing that it stopped. Nothing here is a result, and nothing here involves a patient. What it does tell a biologics team is that the design models cleared whatever internal review Novartis runs before outside software touches real programs, and that Chai now has three of the largest drug companies on that list. Treat the names as a signal about the software being taken seriously, not as evidence that it works.
Source reviewed: Chai Discovery ... Company source ... Read the company brief
BrightHeart said on 8 July 2026 that it received a CE mark and is launching its B-Right AI platform across the European Union, with its European debut at the Fetal Medicine Foundation World Congress in Vienna held from 28 June to 2 July 2026. The company states that both B-Right Views and B-Right Screen are Class II FDA cleared and Class IIa CE marked, with notified body 3022.
Why it matters: Two approvals in two systems is the practical difference between a pilot and a product a European hospital group can buy. The two products are cleared and marked separately, so a buyer should check which one covers the job in front of them rather than treating the platform as a single permission. Neither approval says anything about how many heart defects the software will find in a particular clinic, which is a question for the evidence, not the paperwork.
Source reviewed: BrightHeart ... Company source ... Read the company brief
Commure said on 30 June 2026 that it launched Orchestrator, a referral management and patient intake platform aimed at referral leakage. The company says it takes in unstructured referral data, pulls out the clinical and payer details, checks them against business rules, coordinates teams through one dashboard and writes accepted referrals into records systems, across inpatient, outpatient, post-acute, home health, hospice and specialty settings. Commure says it is already live with leading home health and ambulatory health systems and has processed hundreds of thousands of tasks fully autonomously, and quotes its president Dan Warner saying that every referral that does not convert is a patient who fell through the cracks.
Why it matters: Referral intake is still fax and phone work at most organisations, so there is real time to recover here, and writing the accepted referral into the records system is the part that decides whether staff keep using it. The announcement names no customer, which matters more than usual for a launch, because a platform that works in home health may behave very differently in a specialty clinic. A count of autonomous tasks is also not a conversion rate, and conversion is what referral leakage actually means.
Source reviewed: Commure ... Company source ... Read the company brief
Insilico Medicine said on 29 June 2026 that it signed a research collaboration with SK Biopharmaceuticals to find AI-designed drug candidates for neuroimmune disorders of the central nervous system, covering neuroinflammatory, neurodegenerative and rare neurological conditions. The company states a total deal value exceeding 2.5 billion dollars in potential development, regulatory and commercial milestones, with up to 18 million dollars in upfront and near-term milestone payments and single-digit royalties on net sales. Insilico says it uses its Pharma.AI platform for discovery while SK Biopharmaceuticals handles late-stage development and commercialisation, and reports nominating 31 preclinical candidates since 2021 with 13 receiving investigational new drug approval, from a pipeline of more than 40 candidates.
Why it matters: Milestone-based deal values in this industry are almost always a ceiling that is never reached, because each payment depends on a molecule clearing a stage most molecules do not clear. The cash that matters now is up to 18 million dollars, and for a company listed on the Hong Kong exchange that is the figure a shareholder should model. The genuinely interesting number is the other one: 31 preclinical candidates since 2021 with 13 investigational new drug approvals is a real conversion rate, and it is the measure of whether the AI platform works. Neurological disease is also among the hardest places to prove a drug, so expect a long wait before any milestone triggers.
Source reviewed: Insilico Medicine ... Company source ... Read the company brief
Innovaccer said the agreement covers joint investment in product development, market work, and deployment on AWS services.
Why it matters: AWS can give Innovaccer scale, purchasing reach, and familiar infrastructure for health systems. The same relationship can increase dependence on one cloud provider. The deal matters on both sides: what becomes easier to deploy and what becomes harder to move later.
Source reviewed: Innovaccer ... Company source ... Read the company brief
Trase said on 25 June 2026 that it turned on a fax routing agent in Duke Health's Division of Cardiology on a Monday morning in mid-April, with more than 800 faxes waiting, and that the agent had done almost all of them by the end of the day. The company says the division receives 5,000 faxes a month, that inbound fax triage ran 7.1 times faster, that 1,395 hours a month were returned to clinical staff, and that accuracy went from roughly 70 percent before the agent to 98.1 percent after it. Trase says it worked with Dr Manesh Patel, chief of the division, to pick the task, and that more agents are planned for heart failure, hypertension and high cholesterol follow-up.
Why it matters: The honest part of this story is the starting point: hand sorting was running at roughly 70 percent accuracy, so the comparison is not AI against a careful human but AI against a tired one on a Monday. That is where this kind of automation earns its money, and it is also why the benchmark should be audited rather than taken from the vendor. A 98.1 percent routing accuracy still means about 95 misrouted faxes a month at this volume, and a clinic needs a named person to catch those. The chronic condition agents Trase says are next are a much larger step, because follow-up on heart failure is closer to clinical work than sorting paper.
Source reviewed: Trase ... Company source ... Read the company brief
Trase said on 25 June 2026 that it raised a 107 million dollar seed round led by ARCH Venture Partners, with Red Cell Partners taking part, and that it will use the money to expand its sales team and keep building Trase Origin, its agent operating system. The release quotes Dr Manesh Patel, chief of the Division of Cardiology at Duke Health, saying the team used to arrive on Monday mornings to hundreds of faxes waiting to be routed and that the Trase agent processed the lion's share of them by the end of the day. The release reports the Duke fax agent running 7.1 times faster than manual processing, saving 1,395 staff hours a month and unlocking 285,450 dollars a year of staff capacity.
Why it matters: Money this early usually buys a sales team before it buys a support team, and the release says the sales team is exactly what it is for. A health system signing now should ask who answers the phone when an agent breaks, because that is the function a growth round tends to fund last. Trase also sells into defence and other regulated industries, so healthcare is one vertical competing for the same engineers, and the 285,450 dollar figure is one division at one hospital rather than a rate card.
Source reviewed: Trase press release carried by Yahoo Finance ... Company source ... Read the company brief
RadNet said DeepHealth received clearances for breast arterial calcification assessment and prior-exam integration in its breast imaging products.
Why it matters: One clearance adds a cardiovascular signal to a breast imaging visit. The other brings prior exams into the current reading. Both could make screening more useful, but the value depends on the FDA-cleared intended use, false-positive behavior, and how the tools fit into a radiologist's work.
Source reviewed: RadNet ... Public parent source ... Read the company brief
Artera published on June 24, 2026 a report titled Voice AI in Specialty Patient Access, drawn from interviews with specialty group leaders in dermatology, ophthalmology, orthopedics and multispecialty organizations. Artera says many voice AI pilots stall on weak vendor selection rather than weak technology, and describes a demo problem in which vendors show tidy calls while real patient lines carry background noise, upset callers, unclear insurance and odd cases. The report says the groups that do best agree on their success measures before the pilot starts, that many practices hold no baseline numbers for call abandonment or scheduling time, and that vendor claims about records system connections are often larger than what the product does today.
Why it matters: A vendor publishing advice on how to choose a vendor is selling as well as teaching, and the advice still holds up: a practice with no baseline call numbers cannot tell whether a pilot worked. Artera does not say how many leaders it interviewed, so the findings read as themes rather than measurements. The useful move for a practice is to write down its current abandonment rate and scheduling time before the first demo, because that number is the only thing a pilot can be judged against.
Source reviewed: Artera ... Company research ... Read the company brief
Sonio said on 24 June 2026 that the FDA cleared the Sonio Suspect Integration Configuration under K261519, a version of its abnormal fetal finding detection that can be built into an ultrasound machine instead of running only as a standalone web application. The company says it keeps working where computing power is limited, calls it the first AI cleared as an integration module in an ultrasound machine for anomaly detection, and describes it as its fifth FDA clearance in under three years.
Why it matters: Running inside the scanner removes the second screen and the network dependency, which is the part that usually kills imaging AI in a busy clinic or a site with thin bandwidth. It also makes the software something a machine maker can ship, and Samsung owns Sonio, so a buyer should ask which scanners the module will actually reach. The clearance covers this configuration, not a new claim about how well the detection works, so the performance question is still the earlier evidence.
Source reviewed: Sonio ... Company source ... Read the company brief
Hologic announced on June 16, 2026 that the Unified Patent Court in Germany ruled on June 10, 2026 that Siemens Healthineers infringed European Patent EP 2 352 431, which covers focusing technology in the Hologic Envision platform. Hologic says the court granted an injunction against making, selling, marketing, using, importing or storing the MAMMOMAT B.brilliant systems in Germany, France and the Netherlands, ordered a recall and destruction of affected systems in those countries, required Siemens Healthineers to refund customers and cover logistics costs, and left past and future damages for the court to set. Essex Mitchell, chief operating officer at Hologic, gave the company statement. The release names Siemens Healthineers AG, Siemens Healthcare GmbH, Siemens Healthineers Nederland B.V. and Siemens Healthcare SAS as the parties, and does not mention AI.
Why it matters: This is a hardware and patent fight, not an AI story, and that is exactly why it matters to a buyer of imaging AI. Breast imaging AI runs on top of a mammography system, so a hospital in one of these three countries now faces a scanner being removed and a decision about what replaces it, which drags every connected software choice along with it. Only Hologic has published its account here, so the scope of the recall, any appeal, and the final damages are all still open. A buyer outside Germany, France and the Netherlands is not directly affected by this order.
Source reviewed: Hologic ... Company source ... Read the company brief
The company announced a model designed to draft radiology reports from images, prior exams, and clinical context.
Why it matters: Drafting the report changes more of the clinical workflow than flagging one finding. It could save radiologists time, but the signed report still needs accuracy, review, and clear responsibility. A board-exam result is interesting; prospective clinical performance is the stronger receipt.
Source reviewed: Harrison.ai ... Company source ... Read the company brief
Philips reported survey findings on clinician AI use, time savings, capacity, safety, training, and readiness in the United States.
Why it matters: The report gives a current picture of clinician attitudes and claimed time savings. It also comes from a company with a commercial interest in healthcare technology adoption. That does not make the findings useless. It means the survey method and independent comparisons carry more weight.
Source reviewed: Philips ... Company research ... Read the company brief
The company announced a council intended to define a more consistent way to measure coding quality for both people and AI.
Why it matters: Medical coding vendors can all say their accuracy is high when they are not using the same definition or test. A shared standard could make those claims comparable. The council becomes more valuable if its method, membership, conflicts, and results are public enough for others to repeat.
Source reviewed: CodaMetrix ... Company source ... Read the company brief
Owkin announced on 5 June 2026 a five-year licence with Sanofi for K Pro, the product Owkin calls its AI Scientist, together with work to build what it describes as next-generation biopharma agents purpose-built for Sanofi. Owkin says the agents are meant to act as assistants that carry out complex tasks in drug research and development on their own, across the span from early discovery through clinical development, and to support competitive intelligence and decision-making. The announcement discloses no financial terms. Thomas Clozel, Owkin's co-founder and chief executive, is quoted saying that building on the existing collaboration with Sanofi marks a shift toward truly embedded AI. Emmanuel Frenehard, Sanofi's chief digital officer, is quoted saying that putting purpose-built agentic systems into the company's workflows is meant to let its teams work with greater speed, depth and confidence.
Why it matters: The length is the signal. A five-year licence has to survive budget reviews and a change of leadership, which a one-year pilot does not, so Sanofi has committed to living with this software rather than testing it. Two things are worth holding at arm's length. No money was disclosed, so the term tells you about intent and not about scale. And the word autonomous is doing a lot of work: an agent that writes a research summary without being asked is useful, while an agent that prioritises which molecule to pursue is a decision, and the announcement does not separate the two. Note also that Sanofi's chief digital officer is the named executive rather than a research leader, which suggests this started as a digital programme.
Source reviewed: Owkin ... Company source ... Read the company brief
Artera said on 4 June 2026 that Artera Harmony Federal Edition achieved FedRAMP Class D certification, the level formerly called FedRAMP High, and that fewer than 100 companies worldwide hold it. The company says it met more than 400 security controls covering access control, encryption, physical security and incident response, reaching In Process status in August 2025 with Department of Veterans Affairs sponsorship, an Authority to Operate in February 2026 and FedRAMP authorisation in May 2026. It says the certification deepens its work with the Department of Veterans Affairs, the Department of Defense and the Indian Health Service through Artera Government Solutions.
Why it matters: Federal health agencies cannot buy software that has not been through this process, so the certification is less a feature than a licence to bid, and it took Artera about ten months from In Process to authorisation. For a federal buyer that means a shorter list and a faster procurement. For a commercial health system it means something different and smaller: the controls apply to the Federal Edition of the product, not automatically to the version a private hospital would run, so ask which edition a quote covers.
Source reviewed: Artera ... Company source ... Read the company brief
Avicenna.AI published on June 3, 2026 a review of six peer-reviewed studies covering AI detection of pulmonary embolism on CT pulmonary angiography. The company reports that in a real-world study of 1,204 examinations, clot-positive cases not reported at first read fell from 15.6 percent to 3.8 percent, with per-case sensitivity of 93.9 percent and specificity of 94.8 percent. It reports time from scan to specialist consult falling from 241 minutes to 6.7 minutes at one US referral centre, scan-to-alert time falling from 318 minutes to 5.47 minutes in a follow-up at the same institution, and in-hospital deaths falling from 8.4 percent among 113 patients before deployment to 2.2 percent among 45 patients after. In a reader study of 207 examinations, radiology residents reached 92.5 percent sensitivity with the tool against 81.7 percent without it. Avicenna.AI describes the product, CINA-PE, as a CE marked Class IIb decision-support tool with FDA 510(k) clearance K210237. This site's own FDA device records list K210237 as a 510(k) clearance decided on May 19, 2021 under the applicant name Avicenna.Ai, recorded there with the device name CINA CHEST.
Why it matters: Publishing six studies with confidence intervals, rather than one percentage on a slide, is the behaviour a cautious buyer should reward, and the sensitivity figures hold up across several CT vendors and scanner models. The weakest numbers in the set are the mortality ones, because 113 patients before and 45 after is a small single-site comparison where staffing, triage habits and case mix all changed at the same time as the software. The time-to-consult drop from four hours to minutes is the result most likely to repeat elsewhere, since a faster alert does not depend on the model being clever, only on it firing.
Source reviewed: Avicenna.AI ... Company research ... Read the company brief
Commure said on 19 May 2026 that it raised 70 million dollars at a 7 billion dollar post-money valuation, led by General Catalyst with Sequoia Capital, Morgan Stanley and Kirkland and Ellis also taking part. The company says the money goes into scaling its revenue cycle and practice management platform across specialty practices, hospitals and integrated delivery networks, into a shared intelligence layer for its agents covering payer rules, specialty coding and denial patterns, and into AI infrastructure for markets outside the United States. It says it operates inside more than 500 healthcare organisations across more than 3,000 sites of care, serves more than 130 of the largest health systems including HCA Healthcare and Tenet Healthcare, processes tens of billions of dollars of annual payments with more than 85 percent of the work completed without human intervention, and supports tens of millions of appointments a year through its ambient AI products.
Why it matters: Raising only 70 million dollars at a 7 billion dollar valuation is a small round against a large price, which usually means the company did not need the cash and was setting a marker rather than funding operations. The number a buyer should care about is the other one: more than 85 percent of revenue cycle work completed without a person, measured across tens of billions of dollars of payments. If that holds up in a specific contract it is the strongest claim in this group, and if it does not, it is the easiest to disprove. Naming HCA Healthcare and Tenet Healthcare is also a real commitment, because those organisations are large enough to be asked directly.
Source reviewed: Commure ... Company source ... Read the company brief
Viz.ai announced the Viz Pulmonary Suite on May 14, 2026, describing it as a combination of acute and chronic pulmonary workflows with context-aware patient summaries, guideline surfacing and tools for specific conditions. The named conditions are chronic obstructive pulmonary disease, lung nodules and pulmonary embolism. The announcement cites time to pulmonary embolism treatment falling from 1.75 days to 0.56 days, and problem figures including 55 to 65 percent of potential in-network referrals lost to health system leakage, nearly 50 percent of hospitalized COPD patients readmitted within 30 days, and up to 71 percent of clinically significant lung nodules going without appropriate follow up. The announcement states no FDA status for the suite itself.
Why it matters: The interesting move here is that this product is partly about keeping referrals inside the health system, and Viz.ai says so plainly by citing leakage rates alongside clinical ones. That makes it an easier sell to a chief financial officer than a pure triage tool, and it means a buyer should ask whether a recommendation is driven by clinical guidance or by network retention. No FDA status is stated for the suite, so each component's regulatory standing needs checking separately before anyone treats an on-screen prompt as a cleared finding.
Source reviewed: Viz.ai ... Company source ... Read the company brief
Isomorphic Labs announced on May 12, 2026 that it raised 2.1 billion dollars in Series B funding led by Thrive Capital. The round includes existing backers Alphabet and GV alongside new investors MGX, Temasek, CapitalG and the UK Sovereign AI Fund. The company says the money will power its AI drug design engine, scale the business globally and progress its drug candidate pipeline. The announcement discloses the round size but not a valuation or a cumulative total raised.
Why it matters: A round this size for a company with no approved medicine is a bet on the design step, not on a product. The useful thing to watch is not the funding but what it buys: AI drug design is judged by candidates that survive real trials, and the field has a long record of promising molecules that failed in people. Sovereign funds appearing on the list is its own signal, because governments now treat drug design capability as infrastructure rather than as a venture investment.
Source reviewed: Isomorphic Labs ... Company source ... Read the company brief
Artera announced on May 7, 2026 an AI services model built around what it calls AI Service Squads, dedicated teams that build custom agent tools for a provider instead of handing over a product to set up. Artera says it supports 2 billion patient messages a year, works with more than 1,000 healthcare organizations including nearly 300 federally qualified health centers, and reaches more than 200 million patients. The company lists connections to Epic and to several other records systems, and co-founder and chief executive Guillaume de Zwirek said healthcare leaders want one AI partner rather than a string of single-purpose tools.
Why it matters: Selling people next to software is an admission that healthcare AI rarely works straight out of the box, and specialty groups and community clinics are the buyers least likely to have their own AI staff. The reach numbers are Artera own counts of messages and customers, so they describe the size of its network rather than how well any agent performs. The question underneath is what the squad leaves behind: a tool the clinic staff can change later, or a custom build only Artera can maintain.
Source reviewed: Artera ... Company source ... Read the company brief
Iterative Health said on 30 April 2026 that it closed a 77 million dollar Series C led by Intrepid Growth Partners and GV, with new investors EDBI and a family office joining existing investors including Insight Partners and Obvious Ventures. The company describes itself as a healthcare technology and services company accelerating clinical research by combining centralised operations, expert staffing, its own AI technology and trial expertise into a multispecialty research network, and says the money funds expansion into cardiology and obesity beyond gastroenterology and hepatology. It reports site activation twice as fast, cutting startup timelines by up to three months, and enrollment rates three times higher, averaging more than two inflammatory bowel disease patients randomised every business day, and notes that nearly 90 percent of United States trials miss enrollment targets on time. Its AI product is named SKOUT.
Why it matters: The interesting shift is that this company started as endoscopy AI and is now a research site network that happens to own AI, and the funding is explicitly for buying into new therapy areas. For a sponsor that is a real option, because a network with its own staff can be held to an activation date in a way a loose group of sites cannot. It also means the AI claim and the operational claim need separating: three times enrollment could come from the algorithm finding patients or from paying coordinators to work harder, and the announcement does not split them. Two patients randomised a business day is a useful, checkable figure in a disease area where one a week is common.
Source reviewed: Iterative Health ... Company source ... Read the company brief
Hologic said on 20 April 2026 that a retrospective single-centre study by Massachusetts General Hospital in Boston reviewed 239 invasive lobular cancer cases diagnosed over 10 years, of which 195 were found during routine screening and 44 were read as negative at the time. The company says its Genius AI Detection solution identified and correctly localised close to 90 percent of the confirmed invasive lobular cancers, and identified 43 percent of the cases that had been read as negative during routine screening. Hologic notes that invasive lobular cancers make up roughly 10 to 15 percent of all breast cancers, and also discussed its 3DQuorum imaging technology.
Why it matters: This is the harder test for mammography AI, because invasive lobular cancer spreads in a pattern that does not form the dense lump the eye and most algorithms are tuned to find. Catching 43 percent of cases that experienced readers called negative is a meaningful result for a second-reader tool. It is also retrospective and from one hospital, which means the algorithm was shown images already known to contain cancer rather than a normal screening day, so it describes a ceiling rather than everyday performance.
Source reviewed: Hologic ... Company source ... Read the company brief
Anumana said on 8 April 2026 that the FDA cleared its ECG-AI algorithm for cardiac amyloidosis read from a standard 12-lead ECG, and the company calls it the first and only one cleared for that use on a standard 12-lead test. Anumana reports 78.9 percent sensitivity and 91.2 percent specificity in adults, from company data on file covering 25,525 patients at four United States health systems. The company also says the device had earlier received FDA Breakthrough Device Designation.
Why it matters: Hospitals already record ECGs by the million, so a cleared algorithm that reads an existing test costs less to add than a new scanner. The performance figures are the company's own data on file rather than an independent trial, and a sensitivity of 78.9 percent means roughly one in five people with the condition in that study group was not flagged. Clearance means the FDA accepted the submission for this use. It does not show how the algorithm will behave on a different patient mix.
Source reviewed: Anumana ... Company source ... Read the company brief
Hologic filed on 7 April 2026 that funds managed by Blackstone and TPG had completed their acquisition of the company, with stockholders receiving 76 dollars per share in cash plus a non-tradable contingent value right of up to 3 dollars paid in two instalments of up to 1.50 dollars each if the Breast Health business hits global revenue goals in fiscal 2026 and 2027. The filing says the common stock has ceased trading and will be delisted from Nasdaq, names a wholly owned subsidiary of the Abu Dhabi Investment Authority and an affiliate of GIC as minority investors, and says Jose E. Almeida became chief executive on completion, succeeding Stephen MacMillan after more than 12 years. The deal was first announced on 21 October 2025 at an enterprise value of up to 18.3 billion dollars, and stockholders approved it on 5 February 2026.
Why it matters: A health system comparing imaging vendors has been using Hologic's quarterly filings to judge whether the company can keep supporting a 10-year equipment estate. Those filings are gone. The contingent value right is also worth reading: part of the price depends on the Breast Health business hitting revenue goals in fiscal 2026 and 2027, which is the same business that sells the AI software, so the owners have a reason to push sales hard in exactly that window. None of that says the products will get worse, but it does mean a buyer should put support and upgrade commitments in the contract rather than trusting a public track record.
Source reviewed: Hologic SEC filing ... Regulatory record ... Read the company brief
Artera said on 30 March 2026 that Atlantic Health, a not-for-profit system serving New Jersey, Pennsylvania and the New York metropolitan area, deployed its AI agents for colonoscopy outreach to lift screening rates. Reporting Atlantic Health's own data from the first 30 days after launch, it says 43 percent of patients contacted by the AI agent answered the call and confirmed their identity, 39 percent confirmed they would attend their upcoming colonoscopy, 7 percent asked the agent questions about the procedure, and the agent reduced time spent on these calls by 38 percent.
Why it matters: The figure most vendors would lead with is the time saving, but the more interesting one is that 7 percent of patients asked the agent a question about the procedure. Colonoscopy preparation confusion is a real reason people skip the appointment, so an agent that answers questions is doing more than confirming a slot. A 43 percent answer rate also means most calls did not connect, which is normal for outbound health calls and is the number a program should plan staffing around. These are Atlantic Health's results over 30 days at one system, not a general rate.
Source reviewed: Artera ... Company source ... Read the company brief
Anumana said on 28 March 2026 that it received FDA 510(k) clearance for an ECG-AI algorithm that looks for pulmonary hypertension on a standard 12-lead ECG, and the company describes it as the first such algorithm cleared for that test. Anumana reports 73 percent sensitivity and 74.4 percent specificity in adults who came in short of breath, from a multi-centre study of 21,066 patients at five United States health systems. The company says the algorithm was developed using more than 250,000 de-identified patient records from Mayo Clinic.
Why it matters: Pulmonary hypertension is usually confirmed with tests that cost more and take longer than an ECG, so a cleared screening signal on a cheap test is worth a look for a cardiology service line. The reported specificity of 74.4 percent also means a large number of people without the condition would be flagged, and somebody has to work through those results. A buyer should price the follow-up workload, not only the software.
Source reviewed: Anumana ... Company source ... Read the company brief
Sonio said on 25 March 2026 that the third version of Sonio Detect received FDA 510(k) clearance for automatically detecting ultrasound views and anatomical structures and for checking image quality in prenatal exams. The company reports that it detects 14 first trimester views at 88.6 percent sensitivity and 98.1 percent specificity, and 40 second and third trimester views at 90.1 percent sensitivity and 99.3 percent specificity, covering up to 65 percent of the ISUOG and AIUM detailed exam protocols and identifying 58 anatomical structures. The announcement does not give a K number.
Why it matters: Checking that the exam was complete is a quieter job than finding a defect, and it is often the one that saves a clinic money, because an incomplete scan means calling the patient back. Covering 65 percent of a protocol also means a third of it is still on the sonographer, so this is a checklist helper rather than a replacement. The sensitivity and specificity figures are the company's own, and view detection accuracy says nothing about how well the separate abnormal-finding product performs.
Source reviewed: Sonio ... Company source ... Read the company brief
Regard said on 24 March 2026 that it expanded its platform into cardiology and surgical service lines, having previously focused on hospital medicine, which it says covers roughly 50 to 55 percent of patients admitted to hospitals. The company describes the platform as reviewing all the data in the electronic health record to recommend a diagnosis and generate draft documentation in the physician's workflow at the point of care, with evidence attached, before the physician enters the patient's room. It names The Guthrie Clinic as piloting the expanded platform across additional service lines, and says its technology is used by thousands of clinicians across more than 150 hospitals and has recommended millions of diagnoses that would otherwise have been missed.
Why it matters: Moving from hospital medicine into cardiology and surgery roughly doubles the patients a tool like this touches, and that changes the review burden rather than just the licence count, because a surgical chart raises different questions than a general medical one. The claim that millions of diagnoses would otherwise have been missed is the company's own and is not independently verified; a suggested diagnosis is a prompt for a physician, not a finding, and that distinction is the whole safety argument for this category. The Guthrie Clinic is described as piloting, so treat the expansion as in testing rather than proven in the new specialties.
Source reviewed: Regard ... Company source ... Read the company brief
Navina said on 17 March 2026 that Summit Health is deploying its AI tool to help clinicians understand and act on a patient's health status at the point of care. The company says the initial implementation onboarded more than 1,000 providers, coders and quality care team members and was completed in seven weeks, and that the platform plugs into Summit Health's electronic health record to cover pre-visit chart review, diagnosis, closing care gaps and automated documentation. Navina says that within the first month clinicians reduced administrative work and consistently reviewed its AI recommendations where appropriate.
Why it matters: A seven-week rollout to more than 1,000 people is the useful number, because the usual failure in this category is not accuracy but a deployment that stalls in one clinic. Onboarding coders and quality staff alongside clinicians also tells you what the tool is really for: this sits where documentation meets risk coding, so the business case runs through accurate diagnosis capture as much as through clinician time. What the announcement does not give is any measured result, only that recommendations were reviewed where appropriate, which is adoption rather than outcome.
Source reviewed: Navina ... Company source ... Read the company brief
Brainomix said on 11 March 2026 that it deployed Brainomix 360 Stroke across the West Virginia University Health System, covering all 25 hospitals in the network, and describes it as the first time the entire WVU network has run on a single unified AI solution. The company says the aim is faster acute stroke decisions and more consistent care across rural and urban facilities. It also cites a Lancet Digital Health paper from 2025 reporting a 100 percent increase in thrombectomy rates across 452,000 patients at 107 hospitals over five years.
Why it matters: Standardising on one platform across a network matters more in stroke than in most specialties, because the decision to transfer a patient depends on everyone reading the scan the same way, and a rural hospital with no neurointerventional team has to trust what the city hospital will do with its images. The published Lancet Digital Health figure is the strongest evidence on this profile and it is independent of this deployment, covering a different set of 107 hospitals, so treat it as context rather than a prediction for West Virginia. A doubling of thrombectomy rates also reflects a whole care pathway changing, not software alone.
Source reviewed: Brainomix ... Company source ... Read the company brief
The company presented AI and digital products across its enterprise healthcare portfolio at HIMSS 2026.
Why it matters: GE HealthCare offers public-market exposure to a large healthcare technology business that includes AI. It is not a pure healthcare AI stock. Its AI exposure becomes more meaningful when orders, revenue, and customer results can be tied to the products on display.
Source reviewed: GE HealthCare Investor Relations ... Investor relations ... Read the company brief
AKASA announced recognition in Forbes’ 2026 startup employer list and described its remote-friendly United States workforce.
Why it matters: An employer award can help candidates find a company they may not have considered. It does not explain the current jobs, location rules, team structure, or qualifications behind a specific role. Those details are where the recognition becomes useful career research.
Source reviewed: AKASA ... Company source ... Read the company brief
PathAI announced on March 3, 2026 that the FDA granted Breakthrough Device Designation to PathAssist Derm, which the company describes as AI designed to analyze digital whole slide images of skin lesions and aid pathologists in their review. The designation is a program status intended to speed development and review, and it is not a clearance or an approval. PathAI's own footnote on the announcement states that PathAssist Derm is for research use only and not for use in diagnostic procedures. The product is therefore not available for diagnostic sale in the United States at the time of the announcement.
Why it matters: This is the regulatory word that gets misread most often in healthcare AI marketing, so it is worth being exact. A Breakthrough Device Designation means the FDA agreed the product addresses a serious condition and deserves faster attention. It says nothing about whether the software works well enough to be cleared, and PathAI is clear in its own footnote that this one cannot be used for diagnosis yet. For a pathology group, the useful date is the day a clearance arrives, not this one.
Source reviewed: PathAI ... Company source ... Read the company brief
Avicenna.AI said on 2 March 2026 that it entered a strategic partnership with Bracco Imaging on computer-aided detection and diagnosis in brain CT angiography. The companies say the work is progressing through development and clinical validation at selected sites in the United States and Europe, with Avicenna.AI contributing image interpretation AI and Bracco Imaging extending its contrast-enhanced CT capability. The announcement states that an FDA submission for the new AI applications is foreseen in the second half of 2026, and names no product.
Why it matters: The honest status here is pre-submission, which is the distinction a buyer should hold onto: a planned FDA filing in the second half of 2026 is not a clearance, and a clearance is not a given. The pairing itself is the interesting part, because a contrast agent company has a commercial reason to make the scan easier to read and already sells into the departments Avicenna.AI wants. For Avicenna.AI, whose seven existing clearances are all for triage and lesion software, moving into detection and diagnosis on contrast-enhanced CT is a step up in regulatory difficulty, not a product extension.
Source reviewed: Avicenna.AI ... Company source ... Read the company brief
Qure.ai said on 26 February 2026 that the FDA granted 510(k) Class II clearance to qXR-Detect, software that finds and sorts chest X-ray findings across six areas: lung, pleura, mediastinum and hila with the heart, bone, hardware, and other findings. The company says the six new indications bring its total to 26 FDA-cleared indications across nine products.
Why it matters: Most radiology AI is bought one finding at a time, so a single cleared product that reports across six areas of the same image can replace several contracts and several integrations. The count is a measure of regulatory paperwork, not of accuracy, and a buyer still has to see the performance data for each finding separately. A tool cleared to detect is also not the same as a tool cleared to triage, so the workflow it slots into matters as much as the list of indications.
Source reviewed: Qure.ai ... Company source ... Read the company brief
Brainomix said on 24 February 2026 that it extended its Series C financing to 25.4 million dollars, or 18.8 million pounds, led by existing investors Parkwalk and Hostplus with the new United States investor Modi Ventures taking part. The company says the money will improve customer support and speed up product work across its Brainomix 360 Stroke and e-Lung platforms. It cites a Lancet Digital Health paper from 2025 reporting a 100 percent increase in thrombectomy rates across 452,000 patients at 107 hospitals over five years.
Why it matters: Bringing in a United States investor alongside the existing British and Australian backers is the signal in this round, because an expansion into American hospitals needs local relationships more than it needs cash. The sum is modest next to the hundreds of millions raised by American imaging AI companies, which means Brainomix has to win on evidence rather than on sales coverage, and the Lancet paper is the evidence it leans on. A hospital signing during a build-out phase should pin down support hours in its own time zone rather than assume them.
Source reviewed: Brainomix ... Company source ... Read the company brief
PathAI announced on February 23, 2026 that Labcorp is expanding their collaboration to deploy AISight Dx across Labcorp's anatomic pathology labs and hospital collaborations, covering what the announcement calls its national network. The platform handles case management, slide review, collaboration and annotation in a digital workflow. The announcement states that AISight Dx is FDA-cleared for primary diagnosis in the United States and is CE-IVD marked for primary diagnosis in the European Economic Area, the United Kingdom and Switzerland. No figures are given for the number of sites, pathologists or case volume involved.
Why it matters: Scale is the hard part of digital pathology, because scanning every slide costs money long before AI saves any. A national reference lab committing its own anatomic pathology network is a stronger signal than a pilot at a single academic center. The missing number is the one that would settle it: without site counts or case volume, nationwide is a description of ambition rather than a measure of deployment.
Source reviewed: PathAI ... Company source ... Read the company brief
BrightHeart said on 6 February 2026 that its B-Right AI platform is available through GE HealthCare's Voluson Solution Store. The company says the collaboration is meant to improve early detection of congenital defects and reduce uncertainty in reading a prenatal scan. It repeats that B-Right Views and B-Right Screen are each Class II FDA cleared and Class IIa CE marked under notified body 3022.
Why it matters: A scanner maker's own store is the cheapest way for a small vendor to reach the people already using that machine, and the Voluson line is widely used in obstetric ultrasound. It also makes the software easier to buy than to compare, so a department should still ask for the evidence rather than treating a listing as a recommendation. GE HealthCare distributing a product is not GE HealthCare validating it, and the clearances remain BrightHeart's.
Source reviewed: BrightHeart ... Company source ... Read the company brief
Brainomix announced on February 2, 2026 the launch of Brainomix 360 Stroke Next Generation at the International Stroke Conference. The company says the release carries an industry-first net water uptake biomarker, which it describes as giving clinicians a deeper view of how severe a stroke injury is directly from a non-contrast CT scan rather than from advanced imaging. Brainomix says the platform won a Red Dot Design Award for usability and innovation, and that more than 60 peer-reviewed studies have examined Brainomix 360.
Why it matters: Reading injury severity from a plain CT matters because a plain CT is the scan a small hospital actually has at 3am, while advanced imaging often means moving the patient somewhere else first. Brainomix presents net water uptake as something that helps a clinician judge severity, which is a long way from proof that it changes a treatment decision, and the launch page does not say which United States regulatory status covers this release or which finding it covers. A design award means the screen is easier to read, and nothing more than that.
Source reviewed: Brainomix ... Company source ... Read the company brief
ScreenPoint Medical posted on 30 January 2026 that the final results of the MASAI trial, short for Mammography Screening with Artificial Intelligence, had been published in The Lancet. The company says the trial randomly assigned more than 105,000 women to screening supported by its Transpara software or to standard double reading, and was led by Dr Kristina Lang at Lund University. The figures ScreenPoint reports from the publication are a 12 percent lower rate of interval cancers, meaning cancers found between screening rounds, 27 percent fewer of the more aggressive non-luminal A interval cancers, a 29 percent higher cancer detection rate, 16 percent fewer invasive cancers and 21 percent fewer large T2 or greater cancers in the AI-supported group, sensitivity of 80.5 percent against 73.8 percent, and a 44 percent lower screen-reading workload. Dr Lang is quoted saying earlier detection may make a meaningful difference for women and their health, and chief executive Pieter Kroese is quoted saying that at this level of evidence the question shifts from whether AI should be used to making sure women everywhere benefit. This site's own FDA device records list Transpara clearance K241831, decided on 25 November 2024, to applicant Screenpoint Medical B.V. ScreenPoint's page additionally names 510(k) K232096 and CE marking 0344, which are the company's own statements rather than entries in this site's records.
Why it matters: A randomised trial is a different class of evidence from the retrospective reviews most imaging AI companies publish, because nobody chose which women went into which group. That is why this one is worth reading closely rather than filing with the rest. Two cautions go with it. The trial ran inside a screening program that uses two radiologists on every scan as standard, so the workload saving will not transfer unchanged to a program with one reader, and neither will the comparison. And the interval cancer figure, which is the number that matters most for whether screening is catching disease early enough, is a 12 percent relative reduction, which sounds larger in a headline than it will look in one program's annual counts. The results do not say Transpara will produce the same figures in another population, and a buyer should ask about reader staffing before using these numbers in a business case.
Source reviewed: ScreenPoint Medical ... Company source ... Read the company brief
BrightHeart said on 9 January 2026 that it raised 11 million euros in a Series A round co-led by Odyssee Venture and GO Capital. The company names Mussallem CHD Alliance, Lift Value, the IDAHO HealthTech Club through Side Angels, founding investor Sofinnova Partners, Professor Laurent Salomon, Sacha Loiseau and John Gridley among those taking part. The announcement says the money is for scaling the B-Right platform and does not break down how it will be spent.
Why it matters: A Series A of this size buys a few years, not a decade, which is the honest way for a hospital to read it when signing a multi-year workflow. The investor list leans European, and the round landed six months before the CE mark, so the funding and the European launch read as one plan rather than two events. For a candidate, the useful detail is that the company is still small enough that a single hire changes the roadmap.
Source reviewed: BrightHeart ... Company source ... Read the company brief
2025
Hologic said on 16 December 2025 that a retrospective analysis by Massachusetts General Hospital in Boston, published in the American Journal of Roentgenology, examined 7,500 digital breast tomosynthesis screening exams carried out between 2016 and 2019. The analysis found 100 false negative cases, and the company says its Genius AI Detection solution marked about one third of them, 32 percent.
Why it matters: A screening program's real cost is the cancer found a year late, so a second reader that catches a third of the misses is worth pricing even if it changes nothing on the other 7,400 exams. The study is retrospective and looks only at cases later known to be cancer, so it cannot tell you how many extra healthy women would have been called back. A program should ask for both halves of that trade before committing, because recall rate is what the staff will feel every day.
Source reviewed: Hologic ... Company source ... Read the company brief
Digital Diagnostics said on 25 November 2025 that The People's Health Clinic in Utah, which serves uninsured residents of Summit and Wasatch counties, is using LumineticsCore to screen patients with diabetes for eye disease. The company says the clinic completed nearly 100 eye exams in about three months and that roughly 25 percent of those patients were identified for urgent referral to an ophthalmologist.
Why it matters: A one in four urgent referral rate is high, and the honest reading is that it says as much about the population as about the software: a clinic serving uninsured patients is likely seeing people whose diabetes has gone unwatched for years. That is exactly where moving the exam into the building changes outcomes, because the barrier was never the reading, it was the second appointment. Nearly 100 exams is a small start, and the figures are the company's own, so a larger clinic should ask for referral rates from sites that look like its own patient mix.
Source reviewed: Digital Diagnostics ... Company source ... Read the company brief
Avicenna.AI said on 13 November 2025 that it launched AVI, a platform that delivers AI results into existing hospital systems. The company says it integrates directly with PACS and RIS without an extra viewer or a change to the workflow, runs fully automatically and is invisible to end users, and simplifies deployment for IT teams. It says the platform initially hosts the CINA portfolio, naming CINA-ICH, CINA-LVO and CINA-ASPECTS for neuro, CINA-VCF, CINA-VCF Quantix and CINA-CSpine for spine, and CINA-AD, CINA-PE and CINA-iPE for chest. Avicenna.AI says AVI supports workflows compliant with GDPR and HIPAA, de-identifying data before processing and re-identifying it on return, with encryption and access controls.
Why it matters: The delivery layer is usually what kills imaging AI projects, not the algorithm, so a vendor shipping its own integration path instead of asking for another workstation is addressing the real obstacle. Listing all nine CINA products on one platform also means a department can add findings without another integration project, which is how this kind of purchase actually grows. Note what the platform is not: AVI is plumbing, and each CINA product carries its own clearance, so the platform does not extend any product's permitted use. The de-identification step is worth checking in detail, since it is the company's description of its own design rather than an audited control.
Source reviewed: Avicenna.AI ... Company source ... Read the company brief
The company said its merger resolutions passed, the prior entity ceased to exist, and its shares would be delisted the next day.
Why it matters: The merger means BenevolentAI should no longer be presented as a current public stock. Older pages can keep circulating after a company changes structure, which makes the date of every ownership and market claim important. This update is the line between the former listing and the company that exists now.
Source reviewed: BenevolentAI ... Company source ... Read the company brief
The source records the announcement. The brief adds context and the way we would research it. Inclusion means a development is worth tracking, not that a company, security, course, or job is a recommendation.