Radiology & Imaging AI

Mindray TE Air is indexed as a product, on the same basis as Vscan Air and Philips Lumify, and it sits closer to the membership boundary of this index than any other record in the handheld cluster. That is worth stating at the top rather than burying, because a reader comparing grades should know that this device is admitted for a thinner set of models than its neighbours carry. The series has two wireless probes. The e5M is a two in one whole body scanner combining convex and linear modes in a single housing, with more than 20 presets across abdomen, small organ, vascular, obstetrics, gynaecology and FAST, 22 exam types, scanning depth to 40 cm and roughly 120 minutes of battery. The i3P is a phased array with software based beam steering aimed at emergency and critical care cardiac work, and it is unusual in the cluster for being usable both as a mobile probe and as a transducer on the manufacturer's own trolley system, which turns one purchase into two deployment models. Both are rated IP68 and fully immersible for high level disinfection, both charge from empty to 90 percent in about 35 minutes, and both run through an application on an iOS or Android device the customer already owns. Imaging rests on the manufacturer's eWave platform and second generation single crystal transducer technology. The intelligence is narrow and mostly clerical. Scene recognition detects anatomy and switches and optimises presets automatically between abdomen, thyroid and carotid. iTouch+ performs one click image optimisation. PW Auto Calc traces a pulsed wave spectrum automatically and returns the standard velocity and resistance indices without manual measurement. Smart Bladder computes bladder volume from three automatically recognised diameters. These are real recognition models doing real work, and none of them produces a clinical finding. The richer tool set the manufacturer markets across its wider ultrasound line, covering B line counting, velocity time integral, inferior vena cava, fetal heart rate and automatic ejection fraction, runs on the trolley and portable systems rather than on this handheld, although an i3P docked to the trolley system reaches them. Where this product genuinely leads the cluster is commercial posture. Both probes are sold outright through the manufacturer's own online store with no subscription and an explicit commitment to no recurring fees, a standard three year warranty on the transducer and one year on accessories, published support hours, and a try before you buy demonstration programme. Against neighbours that gate intelligence, storage or imaging interchange behind annual tiers, buying the thing once and owning it is a materially different proposition and the manufacturer says so plainly rather than leaving a buyer to work it out. The manufacturer is headquartered in Shenzhen, China, with North American operations in Mahwah, New Jersey, and has been building medical devices for more than 30 years across patient monitoring, anaesthesia and ultrasound. Two things a reader should weigh. The published evidence base evaluates image quality rather than the models: independent comparative work rates this device well against cart systems and against other handhelds, and no validation of the recognition or measurement features was located, with the 2024 clearance summary recording that clinical studies were not required to support substantial equivalence. And some of the smart tools are sold as one time optional upgrades whose prices could not be located, so the no fees claim holds for the base device and is not the whole picture.

Philips Lumify is indexed as a product, on the same basis as Vscan Air. Royal Philips is far too broad to grade against these axes, so the assessment covers Lumify and the software that runs on it rather than the parent's wider portfolio. The form factor is the oldest idea in this cluster and still the most frugal. Lumify is a cabled transducer that plugs into a smart device the customer already owns, over USB on Android or Lightning on Apple hardware, and the app is downloaded from the ordinary consumer stores. There is no proprietary tablet, no wireless pairing and no battery in the probe, which is why scanning runs to roughly 7.5 hours on a suitable Android device where wireless competitors are measured in tens of minutes. Three broadband transducers cover the range: S4-1 phased array, C5-2 curved array and L12-4 linear. Fourteen presets include abdominal, cardiac, lung, FAST, musculoskeletal, obstetric and, more recently, ocular. Pulse wave Doppler arrived in 2022, adding hemodynamic quantification to what had been a 2D and colour device. Tissue Harmonic Imaging, SonoCT and xRes carry over from the cart line. Lumify was introduced in 2015. Two artificial intelligence features ship, both narrow and both branded. Auto B-line Quantification counts B lines automatically frame by frame during a lung exam and reports the maximum per zone against a guided 12 point protocol with zone by zone labelling. It was cleared in two stages, a counting algorithm first and then automated merged B lines in June 2023 across all three transducers. Auto EF calculates left ventricular ejection fraction from a single apical four chamber view with no manual editing, positioned explicitly as standardising quantification across users of varying experience. The regulatory paperwork is more revealing than the marketing, which is unusual and to the vendor's credit. The clearance summary for Auto EF identifies the underlying model as the LVivo EF algorithm developed by DiA, naming the outside supplier in a public document. The B lines summaries describe the counting mechanism, the dataset construction across lung zones and the ground truth method, which was agreement against a clinician majority with a pre defined success threshold. The lung work was developed under a partnership with the United States Biomedical Advanced Research and Development Authority, and the 2023 clearance was counted as the 74th authorisation supported under that programme, so a buyer can see that public money helped build the feature. Tele ultrasound runs through Reacts, a platform from Innovative Imaging Technologies integrated so a remote colleague can see the live scan and the device camera at once and the clinician controls what is shared. The device turns up in genuinely austere settings, including a Philips Foundation programme in Kenya training village midwives to perform antenatal screening with radiologist support at a distance. Three things a reader should weigh. Both artificial intelligence features are Android only, so a buyer standardised on Apple hardware gets an excellent scanner with no models on it, and that restriction is not prominent in the marketing. Pricing is the least legible in this cluster, with no published United States figure and third party signals that disagree by roughly a factor of two. And the published evidence for the two shipped models is thinner than for the cluster leaders, resting mainly on the clearance summaries rather than on independent trials.

Vscan Air is indexed as a product rather than as a company. GE HealthCare is far too broad to grade against these axes, so this record follows the Optum Integrity One precedent: the qualifying product line is assessed on its own terms and the parent is named. Everything below concerns Vscan Air and the software sold on it, not GE HealthCare's wider portfolio. The device is a wireless dual probe that pairs to a phone or tablet and fits in a coat pocket. Two configurations exist. The CL couples a convex and a linear transducer for abdominal, obstetric and deeper imaging. The SL couples a sector phased array and a linear transducer for cardiac, lung, vascular access and musculoskeletal work. B mode, M mode, colour Doppler and pulsed wave Doppler ship with the probe and function without any subscription, the probe is built to survive drops and rain, and scan time runs to roughly 50 minutes. Vscan Air launched in 2021, the current generation of a handheld line GE began with Vscan in 2010. The intelligence arrived by acquisition. GE HealthCare bought Caption Health in 2023 and brought Caption AI to Vscan Air SL in April 2024. It has two parts. Caption Guidance gives real time prescriptive instruction on how to manoeuvre the probe toward standard echocardiographic views and captures automatically once quality is reached. Caption Interpretation AutoEF calculates left ventricular ejection fraction from one or any combination of three views without manual endocardial border tracing, and the vendor states the calculation grows more accurate as each additional view is acquired. That software carries the strongest regulatory credential in this cluster. Caption Guidance was authorised through the De Novo pathway in February 2020 as the first artificial intelligence permitted to guide the acquisition of medical images, which means it created a device classification rather than matching an existing predicate. A separate expedited clearance followed in May 2020 for frontline pandemic use. The pivotal evidence is correspondingly strong and it is public. Published in JAMA Cardiology in 2021 from a registered trial, eight nurses with no prior ultrasound experience each scanned 30 patients across two hospitals, paired against sonographers using the same hardware without the software, with five independently trained echocardiographers blindly evaluating every acquisition. Images were of diagnostic quality for left ventricular size and function in 98.8 percent of patients, right ventricular size and function in 92.5 percent, and pericardial effusion in 98.8 percent, with at least 92.5 percent agreement on the corresponding diagnostic assessments. Results were reported as consistent across body mass index, sex and race, which is subgroup reporting almost no vendor in this index publishes. Several authors were employees of the company whose product was under test, and that is recorded here rather than passed on unqualified. Three things a reader should weigh. The intelligence is an option rather than a property of the device: it is cardiac only, sold separately, available on the SL and not the CL, and explicitly not available in every country. Interoperability follows the same pattern, with imaging interchange and cloud storage sitting behind a separate paid tier rather than shipping with the probe. And the published terms state that GE HealthCare may collect, derive from and use non personal data related to its products and services for research, development and product improvement, and that it owns the resulting rights while undertaking not to sell the data or identify the customer without consent. That clause is disclosed rather than buried, which is to its credit, and what counts as non personal data is not defined anywhere located.

EchoNous sits between the two poles of this cluster. Butterfly Network puts a single semiconductor probe against the whole body and Clarius sells a range of purpose tuned wireless scanners. EchoNous built Kosmos to argue that a handheld need not concede performance to a cart at all, and the claim rests on hardware rather than software: proprietary piezoelectric transducers, proprietary application specific circuits, and continuous wave Doppler, which the company states no other handheld offers. Continuous wave Doppler is what allows valve gradients and pressure calculations at the bedside, and it is the reason the device is positioned against cart systems rather than against pocket scanners. The artificial intelligence is a set of workflow models rather than a diagnostic one. AI FAST labels anatomy and identifies views in real time, an assisted ejection fraction workflow produces an automated left ventricular measurement, Auto Preset switches the exam preset as the operator moves between abdomen, lung and heart, and Auto Doppler positions the sample gate on the valve the clinician wants to interrogate. A February 2026 release added cleared obstetric biometry and billable vascular workflows. A June 2026 release added Vascular Auto Flow for fistula monitoring in dialysis patients, tricuspid valve Doppler calculations, and support for DICOMweb. Two third party models are offered on the platform and named openly: Us2.ai for automated echocardiography reports and 19Labs for tele ultrasound streaming. The evidence base is the strongest in this part of the index and most of it is not the company's own. Independent groups have validated the automated ejection fraction against cart based echocardiography in a multicentre Japanese study, against cardiac magnetic resonance in an Italian single centre cohort, in a diagnostic accuracy study published in npj Digital Medicine, and in a 115 patient cardio oncology cohort where staff without ultrasound expertise reached accuracies of 89 to 94 percent. The independent work also reports against the product's interest, the Japanese study finding that the automated system tended to underestimate ventricular volumes. The company separately publishes a prospective 78 patient study adjudicated blind by echocardiography core laboratories at two academic centres, which is well constructed and is the vendor's own. Data handling is architectural rather than contractual, and unusually clear. The products privacy policy states that protected health information is accessed only locally on the device, that only device identifiers and crash logs are transmitted, that the transmitted data contains no protected health information, and that EchoNous does not access, collect or disclose patient data. A Chief Privacy Officer is named. Scanning works offline, and a network is needed only to register a transducer, export to an imaging destination, or update software. The company was founded in Redmond, Washington by Kevin Goodwin, who had earlier founded SonoSite and shipped the first point of care ultrasound device in 1999, and by Dr Niko Pagoulatos. It began as Signostics, a bladder scanner business, and renamed itself EchoNous, meaning intelligent sound. Its own account dates the founding to 2016, while its corporate registration and site footer point to 2015. Privately held, backed principally by KKR with later financing from Kennedy Lewis, roughly $194M raised across seven rounds. Kosmos is sold in more than 70 countries and carries a five year warranty. Two things a reader should weigh. The published security posture is the weakest in this cluster, with no audited certification, trust centre or vulnerability disclosure policy located for a networked device sold into hospitals. And commercial disclosure is thin against its own competitors, with a single configuration ceiling published and no statement of what recurring cost, if any, a buyer carries.

Exo, pronounced echo, is the third handheld ultrasound approach in this index and the one that leans hardest on the software half. The company describes itself as a medical imaging software and devices company, in that order, and its silicon takes a middle path between the two competitors already graded here: a piezoelectric micromachined transducer built on silicon, blending the imaging behaviour of piezoelectric crystals with the cost structure of a chip. The Exo Iris device received clearance in 2021 with imaging modes and indications added the following year, and launched commercially in September 2023 at a starting price of 3,500 dollars. The artificial intelligence portfolio is the deepest in this cluster. By June 2025 the company reported 14 cleared indicators embedded in the device, spanning heart failure including reduced ejection fraction, cardiac hypertrophy, cardiomyopathies, valvular disease, pneumonia and pulmonary embolism, alongside bladder, hip and thyroid applications. Two things separate this from measurement automation. The indicators detect findings rather than only measuring structures, and in June 2025 the company cleared what it described as the first artificial intelligence for detecting pleural effusion and consolidation or atelectasis, stating that it showed superiority to clinicians alone. Separately, SweepAI evaluates image quality continuously while the operator sweeps and signals when enough diagnostic information has been captured, and is itself cleared for cardiac and lung use. Exo Works is the workflow half, launched in March 2022, handling documentation, billing, quality assurance and education for point of care ultrasound programmes. Its distinguishing property is that it is not tied to the company's own hardware: it is described as integrating with nearly any point of care ultrasound device and with common hospital record and imaging archive systems, reachable from a phone, tablet or browser. Wellstar MCG Health, formerly Augusta University Health, implemented it across its health system. Security disclosure is the strongest in this cluster and among the strongest in the index. Exo Works earned risk based two year certification under the healthcare assurance framework and completed an audited report at the operating effectiveness level in 2023, the company operates a public trust centre naming those alongside information security management certification, and it states that business continuity plans are validated through those same external audits. Roughly 428 million dollars raised across multiple rounds, most recently in May 2025, with about 192 employees as of March 2026. Led by co founder and chief executive Sandeep Akkaraju, headquartered in Santa Clara, California. Two things a reader should weigh. A vendor stating that its detection model outperforms clinicians alone is making the strongest comparative claim in this category, and no published study, accuracy figure or effect size supporting it was located outside the regulatory process. And the recurring commercial terms are undisclosed: a starting device price is published and nothing describing subscription, warranty or workflow software pricing was found.

Clarius Mobile Health takes the opposite approach to Butterfly Network on the same problem. Where Butterfly puts one semiconductor probe against the whole body, Clarius sells a range of wireless scanners each tuned to a purpose, six models in the current high definition third generation line covering full body, high frequency musculoskeletal and small parts, curved abdominal and obstetric, a dual array model, and an endocavity scanner for urology and women's health. Each connects over its own wireless link to an iOS or Android device, weighs between roughly 280 and 330 grams, and is sealed for full immersion during disinfection. Knobs and buttons were replaced with touch controls, voice control and automation. The artificial intelligence is a model portfolio rather than a feature. The company received clearance in February 2023 for what it described as the first artificial intelligence application cleared for musculoskeletal ultrasound, automatically identifying and measuring tendons in the foot, ankle and knee. In June 2024 it cleared a fetal biometry tool as its eighth model, automatically placing calipers and estimating fetal age, weight and growth intervals, developed on more than 30,000 de identified fetal ultrasound images and aimed explicitly at midwives, nurses and other new ultrasound users in resource limited settings. A prostate model is available in approved regions only. Measurements produced by these tools appear on screen and can be added to the patient record from the application. Commercial disclosure is the most complete encountered in this index. Scanners are priced publicly from roughly 2,995 dollars, membership is listed at 595 dollars a year with a lifetime licence alternative that varies by region, a three year warranty extendable to five is stated, a sixty day return policy is offered and accessory prices are published. Crucially the probe continues to function without a membership; the membership unlocks cloud and local study storage, advanced models, teleguidance and imaging interchange in the standard medical format. That is a materially different bargain from a device that stops saving studies when a subscription lapses. Technical disclosure extends to security parameters, with transport encryption and pairing encryption specified on the product pages alongside a protected health information removal feature and a statement that such data is stored in a compliant manner. The company reports scanners in use across more than 90 countries and, as of 2023, three million high definition scans performed. Privately held, led by president and chief executive Ohad Arazi, headquartered in Vancouver, British Columbia. Two things a reader should weigh. The specialty scanner strategy delivers depth per indication and multiplies cost for an organisation needing several, which is precisely the comparison a buyer should run against a single whole body probe. And no named institutional deployment with a quantified outcome was located, so adoption is evidenced by scan counts and country reach rather than by customer results.

Butterfly Network replaced the piezoelectric crystals inside an ultrasound probe with a silicon chip, and everything else about the company follows from that. A single handheld device images the whole body where conventional systems need several specialised probes, it connects to a phone or tablet, and because the imaging is done in software on general purpose silicon, new capability can be shipped as a cloud delivered update to devices already in clinicians' hands. The current generation is the iQ3, cleared in January 2024 and carrying a European conformity mark, with the earlier iQ+ still sold at a lower price. The artificial intelligence sits in three distinct layers and they are worth separating. The company's own tools assist acquisition and measurement, and in the first quarter of 2026 it became the first company to receive clearance for a blind sweep tool, which estimates fetal gestational age in under two minutes from a sweep that does not require the operator to know how to obtain the standard views. Compass AI is a separately licensed enterprise workflow platform, relaunched in November 2025, addressing documentation, compliance and billing for point of care ultrasound programmes; the company closed its first seven figure total contract value deal for it in early 2026. Butterfly Garden is a partner ecosystem, at 30 partners as of the first quarter of 2026, through which third party developers deliver their own applications on Butterfly hardware, four of them holding their own clearances and one holding a breakthrough designation. A separate line, Butterfly Embedded, licenses the underlying chip technology to other companies, nine as of April 2026. This record is at company level rather than product level. The index rejected Hologic at company level because a portfolio spanning diagnostics, surgical, skeletal health and mammography could not be graded coherently on these axes, and indexed its imaging software line instead. That reasoning does not transfer here: Butterfly is a single probe platform with one cloud and one software line around it. Commercially it is a small public company. First quarter 2026 revenue was 26.5 million dollars, up 25 percent, split for the first time into 20.8 million of core revenue covering probes, software and services and 5.7 million of embedded licensing. Cash stood at 138 million dollars against a quarterly burn of 12.5 million. Headquartered in Burlington, Massachusetts. Buyers span hospitals and health systems, medical schools running one probe per student programmes including Indiana University School of Medicine, veterinary practices, global health programmes and, from 2026, home and community care. The University of Rochester reported a 116 percent increase in point of care ultrasound revenue after a systemwide deployment. Two things a reader should weigh. Device prices and membership tiers are published, which is rare in this index, but a membership is required to activate the device and lapsing drops the probe to live view only, with no study saving, no advanced modes and no artificial intelligence, so for documented clinical use the subscription is effectively permanent. And imaging interchange in the standard medical format is not included in any membership tier and is purchased separately.

contextflow reads chest computed tomography and shows the radiologist comparable cases rather than only a verdict. ADVANCE Chest CT provides computer aided detection support across suspected lung cancer, interstitial lung disease and chronic obstructive pulmonary disease, combining nodule detection and quantification, nodule tracking across time, quantitative lung tissue analysis, and qualitative assessment of 19 named image patterns. Alongside those findings it surfaces reference cases and differential diagnosis information drawn from its founding technology, a three dimensional image based search engine that retrieves visually similar cases from a reference database. That retrieval design is the company's distinguishing position and it is deliberate. Public material and customer commentary both frame the product against black box artificial intelligence, and showing a radiologist several comparable prior cases is a materially different kind of support from issuing a probability. The findings arrive inside the radiologist's own native viewer rather than in a separate application. The company is a spinout of the Medical University of Vienna and the European KHRESMOI research project, supported by the Technical University of Vienna, founded in July 2016 by Markus Holzer, Georg Langs, René Donner and Allan Hanbury. Roughly 14 million dollars was raised across two rounds including a 6.7 million euro Series A in 2021 and a 1.2 million euro European Commission grant in 2020. An advisory board draws on leadership from the European and International radiology societies. Regulatory standing is European and singular. ADVANCE Chest CT is CE certified under the current medical device regulation, a pathway the company has publicly described as requiring substantial investment in quality management, and which a number of legacy products did not survive. No United States clearance was located and no United States market presence was found. Ownership changed in June 2026. 4DMedical Limited, an Australian listed respiratory imaging company, signed a binding agreement to acquire contextflow for more than 11.4 million euros, gaining a European commercial and technical team, established clinical relationships and a CE marked portfolio already in routine use. Two things follow that a reader should hold together. The acquirer holds United States clearances of its own, so the combination could open a market contextflow could not reach alone. But the same acquirer bought Imbio in 2023 and now describes that company's capabilities as part of its own portfolio rather than as a continuing brand, which is the pattern that ends a separate record. This vendor is indexed on its own because the acquisition is recent and the product remains named and marketed, and brand persistence beyond the transition has not been verified. Disclosure outside the clinical and regulatory story is thin. A dedicated pass located no pricing, no security attestation, no data handling statement, no deployment or hosting description, no customer count and no named deployment scale figure.

icometrix measures the brain. Its icobrain software automatically detects, segments and quantifies brain structures and lesions from magnetic resonance and computed tomography images, then reports those measurements against an age and gender matched reference population so a clinician can see whether a given brain differs from expectation and whether it is changing over time. Longitudinal comparison across scans acquired at different times is patented and is the capability the company was built around. Coverage is unusually wide for imaging artificial intelligence. Modules address multiple sclerosis, dementia and Alzheimer's disease, amyloid related imaging abnormality monitoring, traumatic brain injury, stroke, epilepsy, brain tumours and Parkinson's disease, with a portfolio described as eight regulatory approved solutions. Alongside icobrain sit icobridge, which connects magnetic resonance scanners, picture archiving systems and record systems and handles transfer, report generation and dashboards, and icompanion, a patient facing application for people with multiple sclerosis. The regulatory record runs a decade. First United States clearance came in September 2016, European marking followed under the newer medical device regulation pathway, and in June 2026 icobrain aria was cleared as the first artificial intelligence software to detect, measure and grade amyloid related imaging abnormalities, the serious adverse effect associated with new anti amyloid Alzheimer's therapies. The company describes that clearance as the first computer aided detection and diagnosis solution authorised in neuroradiology, a category held to markedly higher evidence standards than quantification alone. In the United Kingdom the multiple sclerosis measures received a Medtech Innovation Briefing from the national health technology assessment body in April 2022. In the United States the service qualifies for reimbursement under an established procedure code for three dimensional post processing. Ownership changed in 2026: icometrix is now a GE HealthCare company. That is the most consequential fact on this record for a prospective buyer, and it cuts both ways. It brings distribution, capital and an installed scanner base, and it raises the questions any acquisition raises about roadmap independence, pricing under a larger vendor's commercial structure, and whether neutrality toward competing scanner manufacturers persists. The brand, the products and the site remain under their own name, which is why this is indexed as its own record. Founded in 2011 as a spinout in Leuven, Belgium, with United States operations established in the Boston area in 2016 and a further presence in New Jersey. Output is delivered as colour coded segmentations in DICOM format, concise reports carrying normative reference data, and prepopulated reporting templates. Data handling is stated to comply with United States health privacy law and European data protection law, with encrypted transfer, secure storage and controlled access. No pricing of any kind was located, and no customer count, deployment scale figure or named health system reference was found in public material.

Optellum scores whether a lung nodule is cancer. Virtual Nodule Clinic identifies, triages and tracks patients with pulmonary nodules, and its Lung Cancer Prediction AI computes a malignancy risk score from full patterns of three dimensional pixels in an ordinary computed tomography scan, using a convolutional neural network trained as a radiomics based digital biomarker. The clinical problem it addresses is specific and large: roughly two million incidental lung nodules are found each year in the United States and a substantial share of those patients never receive follow up, while small tumours treated early carry survival rates far above late stage disease. The company is building a wider platform, LungOS, on what it describes as the first thorax computed tomography foundation model, announced in August 2025. Regulatory standing spans four jurisdictions. United States clearance came in March 2021, the first for artificial intelligence decision support in lung cancer diagnosis, followed by European medical device regulation marking, United Kingdom marking, and Australian class IIb approval in February 2026. The reimbursement position is what sets this record apart and is rare enough in this index to state plainly. Two temporary procedure codes took effect on 1 July 2022 covering quantitative computed tomography tissue characterisation, one standalone and one alongside a concurrent scan. In the hospital outpatient setting the standalone code sits in a new technology payment classification at a rate of roughly 600 to 700 dollars per use, paid at the full rate without multiple procedure discounting. The company maintains a dedicated reimbursement page explaining the mechanics, and it is candid about the limits: these are temporary category codes carrying no relative value units, claims may need supporting documentation to establish medical necessity, and regional contractors decide case by case. Evidence runs from algorithm validation through to health economics. External validation appeared in Thorax in 2020 and risk stratification accuracy in the American Journal of Respiratory and Critical Care Medicine the same year. Real world impact on time to diagnosis and follow up rates was published in BMJ Open Respiratory Research in December 2025 and presented at a European oncology congress. In March 2026 a Vanderbilt led lifetime payer perspective analysis reported the approach cost effective, and in June 2026 the company was awarded competitive national research funding for a multi site pilot in the English health service. Deployment passed 250 clinical sites and three million scans analysed by June 2026, with named customers including the University of Pennsylvania Health System, Vanderbilt University Medical Center, Oxford University Hospitals and Tanner Health. Security is certified against the 2022 revision of the international information security management standard, achieved in August 2024, with the scope enumerated across design, development, manufacture, sales and support, and the company runs a public trust centre. Headquartered at the Oxford Centre for Innovation with a United States office in the Texas Medical Center in Houston. A Series B closed in January 2026 adding new strategic investors.

Brainomix reads brain and lung scans and tells a clinician what it sees, fast enough to change what happens next. Brainomix 360 Stroke interprets acute stroke imaging in real time, and e-Lung quantifies features on CT lung scans to support earlier identification of interstitial lung disease including idiopathic pulmonary fibrosis, built on a proprietary imaging biomarker the company calls the weighted reticulovascular score. The evidence is what distinguishes this record. A study published in The Lancet Digital Health in December 2025 examined 15,377 patients whose scans were reviewed with the tool from January 2022 onward and found clots identified more than an hour earlier. Separately, real world evaluation has been associated with an increase of more than 50 percent in mechanical thrombectomy rates, and the company describes itself as the only stroke imaging tool with demonstrated impact on treatment rates. That is an outcome claim rather than an accuracy claim, and this index has spent a great deal of effort asking other vendors for exactly that distinction. On the lung side, a study with AstraZeneca published in the American Journal of Respiratory and Critical Care Medicine showed the biomarker stratifying patients at risk of pulmonary fibrosis. Regulatory standing is correspondingly strong. The company holds ten United States clearances, eight covering the stroke platform and two covering e-Lung, alongside European marking, with the first stroke clearance in 2023, the e-Lung clearance in May 2024 and a further stroke clearance in April 2025. The deployment and data architecture deserves attention because it answers questions most vendors leave open. Brainomix 360 runs as a managed appliance on a dedicated physical server, on a virtual server using infrastructure the hospital already owns, or in a secure cloud environment. Patient data is stored on the Brainomix server inside the hospital and is not transmitted externally; email notifications carry pseudonymised results only. Data is encrypted in transit and at rest, retention is configured by the customer, and authentication runs through the hospital's own directory using LDAP, Microsoft Active Directory or single sign on with multi factor authentication. The company is certified against the international information security management standard and publishes a trust centre. Results return to the picture archiving system as annotated DICOM series or PDF reports and are also available through web and mobile applications. An Oxford spinout with offices in the United Kingdom, Ireland and Chicago, led by co founder and chief executive Michalis Papadakis. Deployment covers more than 70 hospitals in the English health service alongside sites in the United States and continental Europe. A Series C reached 18.8 million pounds in February 2026 after a 4.8 million pound extension led by Parkwalk Advisors and Hostplus, with United States investor Modi Ventures joining. One gap stands against all of that: no pricing information of any kind was located, and no commercial terms are published anywhere.

Real time autonomous medical coding built on deep learning, sold to providers, payers and revenue cycle management companies rather than to health systems alone. Products are mCoder, the coding engine, and CodePilot, launched November 2024, which surfaces coding intelligence at the point of care rather than after the encounter closes. Headquartered in Tel Aviv. Founding year is unsettled across sources, which give 2017, 2018 and 2019; the company's own about page says 2019 while investor databases cluster on 2018. Founded by Yossi Shahak and Michael Brozino, both former senior McKesson executives. That is an unusual profile in a category dominated by machine learning founders and gives the company an operator rather than researcher orientation. The technical claim rests on proprietary deep learning models plus synthetic data generation, which the company positions as the reason it can reach site specific accuracy without the very large customer chart volumes competitors require for calibration. Stated performance is an 85 percent direct to bill rate at 97 percent accuracy. Direct to bill is the honest metric to compare here, since it measures charts reaching billing untouched rather than accuracy on the subset the engine chose to code. Distribution runs through partnership rather than direct enterprise sales. Maverick completed an implementation at RadNet in December 2024, announced a strategic integration with NewVue.ai and RADPAIR in November 2024, and works with ImagineSoftware. Together these indicate real depth in radiology revenue cycle rather than broad multispecialty coverage, which is why radiology is carried as a secondary category. In August 2025 Infinx made a strategic investment and partnership, which is the relationship most worth watching, since it embeds the engine inside a larger revenue cycle vendor's book of business. Funding is modest at roughly 5.7 million dollars across three rounds from investors including LionBird, Firstime and the Israel Innovation Authority, plus the Infinx corporate investment. Operates in a HIPAA compliant and SOC 2 certified environment. Deployment scale is not publicly disclosed, which is the main gap in this record.

Genius AI Detection is the breast imaging artificial intelligence software line sold by Hologic, Inc. of Marlborough, Massachusetts. It is indexed as a product rather than as a company under the index by product not company rule: Hologic spans diagnostics, gynaecological surgical, skeletal health and mammography hardware, and grading that whole portfolio on capability axes built for clinical artificial intelligence would describe none of it accurately. The record covers the software a breast imaging buyer actually licenses. The line comprises Genius AI Detection, a deep learning detector for digital breast tomosynthesis cleared as K201019 on 18 November 2020; Genius AI Detection 2.0, cleared as K221449 on 6 October 2022 with subsequent clearances K230096 on 23 May 2023 and K243341, which adds automated lesion correlation between craniocaudal and mediolateral oblique views; Genius AI Detection PRO, a combined two dimensional and tomosynthesis platform that incorporates prior examinations, automated pre reporting and image quality checks, with a simplified case score from 1 to 10; Quantra, machine learning breast density assessment; and 3DQuorum, which generates 6 mm SmartSlices from high resolution tomosynthesis data. The predecessor product line, ImageChecker CAD, is two dimensional conventional computer aided detection originating with R2 Technology and is named here because the vendor uses it as the published comparator for its current false positive claims. All clearances sit under product code QDQ, 21 CFR 892.2090. Architecture matters more here than for most records in this lane. The software resides on the acquisition workstation of a Dimensions mammography system and analyses standard and Clarity HD high resolution tomosynthesis, so it reads only images produced by the manufacturer's own hardware. Output travels the other way through open standards, packaged as a DICOM Mammography CAD SR object and distributed for display on DICOM compliant review workstations including SecurView and third party archives. Interoperability is therefore asymmetric: open on the output side, closed on the input side. The underlying tomosynthesis platform is approved under premarket approval P080003, the first tomosynthesis system approved in the United States, granted 11 February 2011 and extended by supplements covering synthesised two dimensional imaging in 2013, the 3Dimensions system in 2018 and high resolution tomosynthesis in 2019. That installed base places this record in an unusual position: Hologic gantries are, in large part, the acquisition hardware that competing algorithms in this lane read, so the company is simultaneously the substrate the lane runs on and a competitor within it. The company's own announcement of 7 April 2026 states that it obtained a ruling in the Unified Patent Court in Germany finding that a competitor infringed a Hologic mammography patent, and that the court enjoined that competitor's mammography system across several European markets. That account is the vendor's own and has not been checked against the court record. Ownership changed materially in 2026. Hologic completed a take private on 7 April 2026, acquired by funds managed by Blackstone and TPG with significant minority investments from ADIA and GIC. Holders received 76 dollars per share in cash plus a non tradable contingent value right of up to 3 dollars in two payments, for total consideration of up to 79 dollars per share, roughly 17.3 billion dollars in cash to equityholders against an enterprise value of up to 18.3 billion. Joe Almeida became chief executive. The contingent value right pays only on achievement of global revenue goals for the Breast Health business specifically in fiscal years 2026 and 2027, which ties part of the seller consideration to the performance of the division this product sits in. The company filed to delist and to terminate its registration, so the periodic financial disclosure that a counterparty could previously rely on has ended. That consequence is graded on the commercial axis; the incentive structure is recorded here as fact rather than as a finding.

DeepHealth is the umbrella brand for RadNet's Digital Health segment and the entity that contracts RadNet's clinical AI and imaging informatics software to third party providers. It is a wholly owned subsidiary of RadNet, Inc. (NASDAQ: RDNT), headquartered in Somerville, Massachusetts, and the brand was launched in November 2023 to consolidate a sequence of acquisitions. Those absorbed brands are named here because this index has no aliases field and they are otherwise liable to be rebuilt as separate vendors: eRAD (radiology information systems and picture archiving, now rebranded into Diagnostic Suite and Operations Suite), the original DeepHealth breast AI company, Aidence (lung), Kheiron (breast), Quantib (prostate and brain), See-Mode (thyroid and vascular ultrasound), iCAD (acquired July 2025 for a reported 103 million dollars, bringing the ProFound Breast Health Suite covering detection, density, risk and breast arterial calcification), CIMAR UK (cloud image management, acquired November 2025) and Gleamer SAS (Paris, acquired 2 March 2026, all cash, up to 230 million euros comprising about 215 million at closing plus a 15 million contingent milestone, bringing musculoskeletal, chest and neuro radiograph AI including BoneView). Gleamer retains its own record at [[gleamer]] because it still carries a distinct brand, product line and customer base. The portfolio is anchored on DeepHealth OS, a cloud native operating system, and divides into enterprise imaging (Diagnostic Suite, a picture archiving replacement; Operations Suite, covering scheduling, registration, billing and analytics; TechLive for remote multimodality scanning; Reporting Pro) and population health clinical AI (Breast Suite, including SmartMammo Dx, also known as Saige-Dx, with density, risk and breast ultrasound modules, plus Chest, Prostate, Neuro and Thyroid suites). Company published scale, combined with Gleamer: more than 2,700 customer contracts across more than 50 countries, 26 FDA cleared and 22 CE marked devices, more than 300 screening sites and more than 10 million mammograms a year. Third party traction is disclosed separately, with more than 16 million dollars of total contract value in new business signed in the first quarter of 2026, and named external customers including ONRAD, which serves more than 120 hospitals and radiology groups, and Wichita Radiological Group. The evidence base is led by the ASSURE study, published in Nature Health in November 2025 (Louis et al., doi 10.1038/s44360-025-00001-0), a prospective analysis of 579,583 women across more than 100 community imaging sites in four states. It compared an AI driven workflow (208,891 examinations) against the prior standard of care (370,692) and reported a 21.6 percent increase in cancer detection rate, 5.6 against 4.6 per thousand, with no significant variation across socio demographic or breast density subgroups and a 22.7 percent gain in women with dense breasts. Co authors include external academic investigators alongside RadNet's chief science officer. The comparison is against historical control rather than randomised, and the sites are substantially RadNet's own, which is addressed on the evidence axis.

Ultromics reads a routine echocardiogram and detects disease a human reader cannot see in it. A spin out from the University of Oxford founded in 2017 by chief executive Ross Upton, working with Professor Paul Leeson at Oxford and Dr Patricia Pellikka at Mayo Clinic, it has built the EchoGo platform around the idea of identifying disease directly rather than automating measurements. It holds four Food and Drug Administration clearances and two Breakthrough Device designations. EchoGo Heart Failure, cleared in 2022, aids detection of heart failure with preserved ejection fraction, a common form that is difficult to diagnose. EchoGo Amyloidosis, cleared in November 2024 under K240860, was the first artificial intelligence tool cleared for cardiac amyloidosis and the first device enrolled in the agency's Total Product Lifecycle Advisory Programme to reach marketing authorisation, from a pilot of fifteen breakthrough cardiovascular devices. The amyloidosis model works from a single routinely acquired apical four chamber videoclip, which is materially less input than the usual pathway of electrocardiography, cardiac magnetic resonance, scintigraphy and sometimes biopsy. A multi centre international study in the European Heart Journal, run with Mayo Clinic and investigators at the University of Chicago Medicine across more than 18 sites, validated it on an external cohort of 2,719 patients with a 22 percent prevalence of the disease, reporting an area under the curve of 0.93 with 85 percent sensitivity and 93 percent specificity across more than 12,500 cases. The company separately publishes the figures from clinical evaluation under intended use conditions in its clearance summary, 84.5 percent sensitivity and 89.7 percent specificity. Performance held across all major subtypes and distinguished the disease from conditions that mimic it, including hypertensive heart disease, heart failure with preserved ejection fraction and hypertrophic cardiomyopathy. Evidence for the heart failure product is cited in the 2025 American Society of Echocardiography diastology guidelines, and that product carries Medicare reimbursement through Category III code 0932T for outpatient use and a new technology add on payment for inpatient settings. More than 25 peer reviewed studies support the platform, which is in use at UChicago Medicine, Northwestern and City of Hope. Development collaborators include Pfizer and Janssen Biotech.

ScreenPoint Medical, based in Nijmegen in the Netherlands and led by chief executive Pieter Kroese, makes Transpara, breast artificial intelligence for mammography screening. It holds the strongest trial evidence of any vendor in this index. Transpara gives each examination a malignancy risk score from 1 to 10, calibrated so roughly a tenth of examinations fall in each score band, and adds marks on suspicious findings with regional risk scores for intermediate and high risk cases. The top one percent of risk, above a threshold of 9.8, is flagged in the image worklist as extra high risk. In the screening workflow that score decides how much human reading an examination receives: low and intermediate risk go to a single reader, high risk to two. That workflow was tested in a randomised controlled trial rather than asserted. The Mammography Screening with Artificial Intelligence trial ran inside the Swedish national screening programme across four sites in the southwest, randomising more than 105,000 women one to one between artificial intelligence supported screening and standard double reading, led by Kristina Lang of Lund University and registered as NCT04838756. The results were published in three stages. A clinical safety analysis in The Lancet Oncology in 2023 reported a 44 percent reduction in screen reading workload. A secondary analysis in The Lancet Digital Health in 2025 reported a 29 percent increase in cancer detection, 338 cancers among 53,043 participants, with no increase in false positives and more small, node negative invasive cancers detected. Final results in The Lancet in January 2026 addressed the endpoint that matters most, cancers appearing between screening rounds: a 12 percent reduction in the interval cancer rate meeting non inferiority, 27 percent fewer aggressive subtypes, sensitivity of 80.5 percent against 73.8 percent, and identical specificity at 98.5 percent. A separate paired non inferiority trial appeared in Nature Medicine in March 2026. Transpara is deployed in more than 30 countries with over 12 million mammograms processed. In April 2026 the company raised 14 million dollars from Insight Partners and Siemens Healthineers alongside 2 million in research grants.

Cerebriu is a Copenhagen company founded in 2018 whose Apollo suite intervenes in a brain MRI examination while it is still happening, rather than analysing the images afterwards. Its founders are Akshay Pai, Robert Lauritzen, who is chief executive, Erik Dam, Mads Nielsen and Martin Lillholm. The suite has three parts. Smart Protocol analyses an abbreviated opening set of images and suggests which sequences should be acquired next, so the protocol adapts to what has already been seen while the patient is still on the table. Smart Alert detects critical findings during the examination itself, naming infarcts, intracranial haemorrhage and tumours, so a scan can be escalated before the patient leaves. Smart Reading prioritises high risk studies in the reporting queue. Regulatory position is European rather than American. Apollo is CE marked under both the older medical device directive and the current regulation, while clearance from the Food and Drug Administration is described as pending and the United States status is stated as for investigation only. In 2025 the company secured CE marking for Apollo Smart Protocol as a product embedded inside a Siemens Healthineers MRI system, which it describes as the first clearance anywhere for clinical artificial intelligence embedded in a scanner from a major manufacturer. A validation study published in the European Journal of Radiology in September 2023 reported roughly 89 percent sensitivity and 90 percent specificity for ischaemic lesions across 800 patient scans. Deployments span hospitals in Denmark, the United States, Norway, India, Israel and Brazil. Funding totals about 26.8 million euros, comprising a 17.4 million euro Series A in 2024 led by North Ventures with EIFO, Denmark's export and investment fund, and Sagitta Ventures, followed by 9.4 million euros in 2025. The company also distributes through two independent radiology artificial intelligence platforms, CARPL and deepc, alongside its manufacturer route.

RapidAI analyses brain and vascular imaging in real time to support urgent treatment decisions, principally in stroke. It was founded in 2012, is based in San Mateo, California, employs between 200 and 500 people and is led by chief executive Karim Karti. Its platform is used in more than 2,500 hospitals across over 100 countries and the company states more than 700 clinical studies stand behind it. The evidence position is the defining feature. Research using its perfusion software contributed to the trials that extended the treatment window in national stroke guidelines, and the company states it holds the only perfusion software with a Food and Drug Administration indication for selecting patients for mechanical thrombectomy, and the only software shown to predict subsequent infarct volume from initial perfusion imaging. It presented 28 scientific abstracts at the International Stroke Conference in February 2026 spanning aneurysm monitoring, ischaemic stroke detection, imaging visualisation and radiology workflow. The product is organised as modules on the Rapid Enterprise Platform, delivered through Rapid Edge Cloud and a radiology application called Rapid Navigator Pro, integrating with existing image archiving and record systems. In November 2025 five further modules were cleared: DeltaFuse, which aligns serial head scans to reveal subtle change and is reported to cut radiologist comparison time by over 36 percent, LMVO for vessel coverage on angiography, MLS for quantifying midline shift, OH for suspected obstructive hydrocephalus, and Aortic for measurement. The company frames this as moving beyond triage notification into characterisation, quantification and tracking change over time. It has raised 75 million dollars, most recently in July 2023, holds a working relationship with Amazon Web Services, and has a partnership with Saudi Arabia's largest healthcare provider covering 20 health clusters.

Eyenuk makes EyeArt, an autonomous artificial intelligence system that grades a retinal photograph and returns a diagnosis without any clinician reading the image. It is one of only three systems cleared by the Food and Drug Administration for diabetic retinopathy testing in the United States, alongside the already indexed Digital Diagnostics and AEYE Health, and it is based in Los Angeles. Clearance came in August 2020 under 510(k) K200667. EyeArt was the first cleared autonomous system able to detect both more than mild and vision threatening diabetic retinopathy in a single test, and the first to return a diagnostic output for each eye separately. Its indication is confined to adults with diabetes not previously diagnosed with more than mild disease. Results are returned at the point of care in under a minute. Two further regulatory positions distinguish it. Version 2.2.0 added clearance for the Topcon NW400 camera alongside two Canon models, making it the only cleared system usable with retinal cameras from different manufacturers rather than tied to one. And version 3.0 holds Class IIb CE marking under the European medical device regulation for three diseases from one exam: diabetic retinopathy including macular edema, age related macular degeneration, and glaucomatous optic nerve damage. It is the only system holding both that three disease European marking and United States clearance for diabetic retinopathy. It is also approved by Health Canada. In September 2025 Norway selected EyeArt as the artificial intelligence system for its national diabetic eye screening programme, following a competitive procurement run by the national purchasing organisation on behalf of the regional health authorities. Grading runs in under 30 seconds with no ophthalmologist review and results are written into the national electronic record. Other deployments include more than 25 centres in Italy, a German diabetes clinic, rural and remote screening in Canada, and Federally Qualified Health Centres in Delaware through a partnership with the American Academy of Ophthalmology. Founded by Kaushal Solanki; chief executive Gaurav Agarwal as of 2025.

Oatmeal Health, founded in May 2022 by Jonathan Govette and based in San Jose, sells lung cancer screening programmes to federally qualified health centres and health plans, serving Medicaid and Medicare populations that screening programmes have historically reached least. Its platform, ICARE, runs four stages: a model over structured and unstructured record data that identifies patients meeting the recommended screening criteria, automated scheduling and reminders, a vision transformer that detects and risk stratifies pulmonary nodules on low dose computed tomography, and a language model chatbot that explains results and next steps in the patient's preferred language and books follow up. A clinical team works alongside the software. Integration is deliberately broad, with named record systems and any imaging vendor or archive. Two things a reader should hold together: the company builds its own models rather than deploying somebody else's, which distinguishes it from several peers in this index, and its own site states that the diagnostic component is not yet cleared by the FDA, with clearance expected in early 2027, while press from early 2025 described a reimbursable Medicare covered diagnostic tool.

Magentiq Eye is a Haifa company founded in 2014 by Dror Zur that builds real time computer aided detection for colonoscopy. Its product, Magentiq-Colo, registered with the FDA as the Magentiq Eye Automatic Polyp Detection System, takes the video feed out of the colonoscope, splits it into frames, analyses them with a deep learning model as the procedure is happening, and returns the video to the monitor with bounding boxes drawn around suspected polyps. The purpose is to reduce the proportion of adenomas an endoscopist walks past. The product carries a CE mark and Israeli approval from mid 2021 and an FDA clearance from July 2023, with an updated cleared version released in January 2026 adding endoscopic tool detection, and a separately CE marked feature set covering polyp size estimation and characterisation plus procedural quality metrics that the United States version does not include. Its evidence is unusually strong for this index: an international multicentre randomised tandem trial across ten centres in Europe, the United States and Israel, published in Lancet Digital Health in February 2024.

Digital Diagnostics, founded in Coralville, Iowa as IDx Technologies by retina specialist and computer engineer Michael Abramoff, MD, PhD, builds autonomous AI diagnostic systems. Its flagship product LumineticsCore, formerly IDx-DR, detects more than mild diabetic retinopathy including macular edema from retinal images and returns a diagnostic report at the point of care without any physician evaluation of the image. In April 2018 it became the first autonomous AI diagnostic system authorised by the FDA in any field of medicine, granted through the De Novo pathway as a Class II device after expedited review under the Breakthrough Devices programme. A 510(k) for the current version cleared in May 2021 on its own predecessor as predicate, carrying the original De Novo special controls forward. The company also markets DermSpot for skin, whose regulatory status was not established in this assessment. LumineticsCore is the reference case for autonomous diagnosis in this index: the model does not assist a clinician, it replaces the interpretation step, and the company states that the creator of the autonomous AI assumes liability for the diagnosis it produces.

TeraRecon, a ConcertAI company, sells two things that are easy to conflate. Intuition is advanced visualisation software with a global installed base the company has reported at roughly 1,900 clinical sites, deliberately independent of any single imaging manufacturer or PACS vendor. Eureka Clinical AI is the SaaS layer on top of it, and its distinguishing property is that it is open to third party algorithms rather than exclusively carrying TeraRecon's own, letting a site manage several vendors' AI through one workflow with PACS integration handled once. Named algorithm partners on Eureka include Riverain Technologies (ClearRead CT chest interpretation, separately indexed here), Imaging Biometrics (neuro oncology MR perfusion), ImageBiopsy Lab (musculoskeletal radiograph measurement), ClariPi (ClariCT.AI, ClariPulmo, ClariSIGMAM) and Coreline Soft (AVIEW LCS+ for lung cancer screening, emphysema quantification and coronary artery calcium scoring). Coverage spans radiology, oncology, cardiology, neurology and vascular surgery. TeraRecon has been a KLAS Category Leader for Advanced Visualization in 2020, 2021 and 2022 and a Best in KLAS provider. It also appears among the eight FDA approved plaque analysis systems benchmarked head to head in the Nature Reviews Cardiology quantitative coronary imaging consensus statement, a comparison that Heartflow, the vendor with the largest evidence base in that field, was unable to supply the required analysis for. This record is scoped to TeraRecon. ConcertAI's wider real world evidence and life sciences business is a different product with a different buyer and is not assessed here.

MEDICAL IP builds AI segmentation and quantification software for medical imaging, and everything else it sells derives from that one capability. MEDIP and MEDIP Pro convert CT and MRI into segmented 3D anatomical models; DeepCatch and the DeepCatch X modules extract quantitative measurements from a single chest X-ray across aorta, heart, torso, tuberculosis and emphysema indications; DeepFore predicts liver cancer occurrence and DeepFore Recur predicts recurrence; TiSepX performs lung volumetry. The same segmentation output feeds downstream products in surgical planning, 3D printed anatomical models (ANATDEL), and augmented and virtual reality anatomy education. Established in September 2015 as the first spin off company from Seoul National University Hospital, led by chief executive Sang Joon Park, with its laboratory inside the SNU clinical research institute and offices in Seoul, San Jose and Xi'an. Two things distinguish the record. The regulatory portfolio is among the broadest in this index, spanning US FDA Class II, European CE with ISO 13485, UK MHRA, China NMPA and Korean MFDS approvals up to Class III. And the evidence is genuinely published rather than asserted: DeepFore appeared in the Journal of Hepatology in May 2025, was featured on that June issue's cover and selected as a Selection of the Month by the editors, which is the strongest single publication credential on this sourcing list. Filed under imaging rather than as surgical decision support. Surgical planning and simulation are applications of the segmentation output, not the product. The anatomy education business is cross listed under workforce and training, where it has its own buyer in medical and dental schools.

MediView XR builds XR90, an augmented reality visualisation and navigation platform for image guided interventional procedures. It renders a patient's own anatomy as a 3D hologram derived from their CT, projects live ultrasound anatomically into the body as the clinician scans, and tracks the path of the needle with a holographic light ray, so the operator can look at the procedural site rather than away at a flat panel monitor. The company describes the effect as X-ray vision; the underlying engineering problem is spatially registering prior CT to live ultrasound and to the instrument in real time. Founded in 2017 in Cleveland on intellectual property developed at the Cleveland Clinic, led by president and chief executive Mina Fahim. XR90 received FDA 510(k) clearance in July 2023 for adjunctive use in minimally invasive ultrasound and CT guided needle based procedures for soft tissue and bone, such as biopsies and tumour ablations. The company states this was the first 510(k) clearance for an augmented reality device combining live imaging with 3D extended reality visualisation across both pre operative and intra operative indications. First clinical use followed in November 2023, a mediastinal lymph node biopsy performed by Dr Bradley B. Pua, Chief of Interventional Radiology at NewYork-Presbyterian / Weill Cornell Medical Center. A trial of the system's navigation accuracy was reported in March 2024. In October 2025 MediView closed a 24 million dollar Series A led by GE HealthCare with participation from Mayo Clinic and Cleveland Clinic. Filed under imaging rather than as surgical decision support: the primary buyer is interventional radiology, the inputs are CT and ultrasound, and the product guides a procedure rather than recommending one.

Calgary based cardiovascular imaging software company whose cvi42 platform is long established across cardiac MRI and CT reading, with cvi42 Plaque receiving FDA 510(k) clearance in October 2025 for AI-enabled coronary plaque analysis quantifying total, calcified and non-calcified plaque. The defining characteristic is deployment architecture, and the company leads with it rather than with accuracy claims: cvi42 Plaque runs on-premise, so coronary artery disease evaluation happens without sending patient data to an external reading service. The CEO frames this directly as giving clinicians greater control over their data, improved study processing times and workflow efficiency, and a Johns Hopkins user describes gaining immediate and interactive control over the analysis. That makes Circle the clearest architectural counterpoint in the cardiovascular CT lane: Heartflow and Elucid both route scans to vendor-side processing, Artrya is cloud-based point-of-care with no external reading teams, and Circle is fully on-premise inside the institution. Buyers with data residency constraints, sovereignty requirements, or objections to imaging leaving the enterprise have effectively one option in this category. The product integrates into existing CT workflows and is stated to be compatible with all major vendor systems, reflecting its heritage as a multi-vendor reading platform rather than a single-algorithm point solution. The company also tracks the reimbursement position closely and describes it accurately, noting Medicare coverage of AI-enabled coronary plaque analysis and the transition from Category III code 0625T with national payment set at $950 to a permanent Category I code effective January 2026, alongside higher base payments for the underlying CCTA exam.

Australian medical technology company, ASX listed as AYA, commercialising the Salix AI cloud platform for near real-time point-of-care assessment of coronary artery disease from CCTA. The platform is modular and each module carries its own regulatory status: Salix Coronary Anatomy was cleared in March 2025, Salix Coronary Plaque received FDA 510(k) clearance in August 2025 under K251837, and Salix Coronary Flow was completed with clinical data validation underway ahead of a further 510(k) submission. The plaque module sits inside the same user interface as the anatomy platform and was enabled in the live product immediately on clearance. The design emphasis is speed and self-sufficiency: assessments in under 10 minutes without switching systems, and explicitly no external reading teams validating scan data, so physicians retain direct control rather than waiting on a vendor-side processing service. That is a genuine architectural difference from Heartflow's analysis-service model. Plaque volume is computed by identifying voxels within target Hounsfield Unit ranges located between the lumen and outer vessel wall, and the product reports plaque features including low attenuation plaque, spotty calcification, positive remodeling and the napkin-ring sign, presented through a personalised 3D heart model. The algorithm was built from thousands of scans contributed by institutions in the US, Australia and Canada. Commercial deployment is live at Tanner Health across five hospitals generating fee-per-scan revenue, with Northeast Georgia Health and Cone Health integrating. The SAPPHIRE study is a retrospective multi-centre real-world study across six major US health systems including Piedmont Healthcare, running in three phases to evaluate the prognostic utility of a proprietary Plaque Dispersion Score, with a stated focus on coronary artery disease in women given that a large majority of women who die of coronary disease had no prior symptoms.

The company that created the non-invasive coronary physiology category and now the broadest platform in the cardiovascular CT lane. Nasdaq listed as HTFL. The Heartflow One platform spans four functions across the CCTA pathway: RoadMap Analysis for rapid stenosis detection, FFRCT Analysis for the physiological significance of each lesion computed from a previously acquired CCTA, Plaque Analysis for quantification and staging, and Planner for intervention planning. FFRCT is the origin product and its distinguishing status is guideline recognition rather than clearance alone: it is supported by the ACC/AHA Chest Pain Guideline, which is a materially higher bar than a 510(k) and something no competitor in this lane holds. Plaque Analysis and RoadMap Analysis were cleared in October 2022, with two further clearances reported in May 2026. Critical regulatory scoping that buyers outside the US must catch: all four functions are cleared for clinical use in the United States, Bahrain, Israel, Saudi Arabia and the UAE, but only FFRCT and Planner are cleared in Europe, the United Kingdom, Australia, Canada and Japan. Plaque Analysis and RoadMap are therefore not available for clinical use in most of the developed world outside the US. The evidence base is the largest in the index for any single vendor: more than 600 peer reviewed publications, over 100 studies assessing more than 365,000 patients, and clinical use across more than 400,000 patients worldwide. Landmark studies include SCOT-HEART, the EMERALD series, the ADVANCE-DK registry with seven year follow up, SMART-CT showing a 25 percent reduction in read times, the PRECISE trial, a Mass General Brigham registry of over 15,000 patients with up to 16 years of follow up, and the DECIDE registry of roughly 22,000 patients across more than 30 US centres. Editorial note worth carrying: when the Nature Reviews Cardiology QCI Study Group benchmarked eight commercially available FDA approved plaque quantification systems head to head, the publication records that Heartflow was unable to provide the required analysis for inclusion. The largest evidence base in the category does not necessarily mean the greatest willingness to be compared directly against competitors, which is the axis on which Lunit earned credit in this index.

Oxford spinout measuring coronary inflammation rather than plaque, from the same routine coronary CT angiography scan. This is what separates it from every other vendor in the cardiovascular CT lane. Coronary inflammation inhibits lipid accumulation in adjacent fat cells, producing a measurable gradient in the perivascular adipose tissue surrounding the artery, and the CaRi-Heart FAI-Score quantifies that gradient as a Fat Attenuation Index. The clinical significance is that it detects risk in patients who look clean on conventional assessment: in the Oxford led ORFAN registry, among patients with no or minimal plaque and zero calcium at CCTA, those with the highest coronary inflammation showed roughly 9.5 to 11 times the cardiac mortality risk and around 5 times the major adverse cardiac event risk compared with those with minimal inflammation. The company operates two distinct products with different regulatory standing, and buyers must not conflate them. CaRi-Plaque, for plaque and stenosis quantification, received FDA 510(k) clearance in March 2025 and sits in the same competitive set as Cleerly, Elucid, HeartFlow and Artrya. CaRi-Heart FAI-Score, the inflammation product and the actual differentiator, is regulatory cleared in the UK, EU and Australia but remains pending FDA clearance for US clinical use, holding only a Category III CPT code approved by the AMA panel. A Category III code is an emerging technology tracking code and is not the Category I code with established payment that applies to automated plaque analysis. Evidence is unusually strong and independent: ORFAN was published in The Lancet as a multicentre longitudinal cohort externally validating the AI-Risk prognostic algorithm in an NHS population, the CRISP-CT study established incremental prognostic value of perivascular FAI beyond traditional risk factors, an ESC Congress 2025 late breaker reported FAI-Score predicting cardiac death beyond hsCRP, and a prospective trial NCT07220304 is registered. Cloud based deployment.

AI analysis of coronary and carotid CT angiography that quantifies and classifies atherosclerotic plaque morphology. PlaqueIQ received FDA 510(k) clearance in October 2024 and is distinguished by its ground truth: the convolutional neural networks were trained on carotid endarterectomy histopathology images co-registered to CTA scans, so plaque components are classified against actual tissue rather than against Hounsfield unit thresholds. Because Hounsfield values overlap between tissue types such as lipid rich necrotic core and intraplaque hemorrhage, the software interprets the spatial distribution of tissue properties across adjacent voxels instead of applying fixed HU ranges. The company markets this as CT Virtual Histology. The product reports patient, vessel, and lesion level plaque composition, with particular emphasis on lipid rich necrotic core, a plaque type associated with heart attack and stroke risk. A carotid artery version launched in October 2025, extending the same approach to stroke risk. An FFRCT product derived from the plaque algorithm, which the company argues yields concordance between plaque findings and ischemia estimates, remains pending FDA clearance and should not be treated as available. Reimbursement is established rather than theoretical: four of the seven Medicare Administrative Contractors extended coverage for AI enabled quantitative coronary plaque analysis from November 2024, the radiology benefit manager EviCore added it to coverage guidelines, and United Healthcare aligned its cardiac imaging guidelines to include coronary CTA plaque quantification across all plans from October 2025. Note that the workflow is not fully autonomous: trained analysts segment the data to create the 3D coronary model before the software classifies and quantifies tissue. Indexed alongside Cleerly, its closest comparator, with which it is routinely benchmarked in the peer reviewed literature.

Chinese origin imaging AI company with US operations, whose InferRead suite spans lung nodule detection and characterization on CT, chest radiography, coronary CTA, and target reconstruction, alongside an InferOperate surgical planning suite covering thorax, liver, and urology. Holds the broadest multi jurisdiction regulatory footprint of any vendor in this lane, with authorizations from the US FDA, European CE, UK UKCA, Japan PMDA, and China NMPA. Its flagship lung CT product was first FDA cleared in 2020 and received further clearance in May 2025 for enhanced features including detection of nodules as small as 4 mm and automated lung density analysis. Deployed for tuberculosis screening programmes including UN procured deployments in Eastern Europe and Central Asia.

French imaging AI company whose Milvue Suite reads plain radiographs across trauma and chest, detecting fractures, dislocations, joint effusions, pneumothorax, opacities, pleural effusions, and nodules, and returning a pre-filled structured report to the RIS or PACS in under 30 seconds alongside roughly 60 automatic measurements. Holds two FDA clearances, TechCare Trauma for musculoskeletal detection and triage plus a chest clearance, with CE Class IIa marking under EU MDR via BSI and Health Canada Class II licensing. Reports analysis of more than 25 million X-rays in 2024 across 600 or more sites spanning Europe, North America, Latin America, the Middle East, and Africa. Offered as both cloud and on-premises deployment, and also distributed through the Blackford imaging platform.

Romanian imaging AI company whose platform analyses chest X-rays, detecting and localizing a reported 148 findings, alongside lung nodule detection on CT with longitudinal comparison across a patient's prior scans to track change over time. Notable for tuberculosis detection specifically, where its performance was assessed in a South African tuberculosis prevalence survey published in The Lancet Digital Health, an unusually independent and public health oriented validation setting for a commercial imaging vendor. CE marked and integrated directly into on-premise or cloud PACS, returning annotated images and structured outputs inside the native radiology viewer, with an integrated reporting engine that converts detections into draft report text for radiologist validation.

UK company whose red dot platform sits at the autonomous end of the imaging AI spectrum: it classifies chest X-rays as High Confidence Normal and auto-reports them, removing those exams from the radiologist worklist entirely, while triaging suspected lung cancer cases for priority reporting, with a companion CT head product. Peer reviewed work in Clinical Radiology reported auto-reporting of 15 percent of chest X-rays as normal at a 0.33 percent error rate, and a separate study reported a 60 percent reduction in missed lung cancer when triage ran alongside consultant radiologists. Important corporate history: the original Behold.ai Technologies Limited entered administration in January 2025 after failing to win contracts under the NHS AI Diagnostic Fund, and NICE withdrew red dot from its recommendations in February 2025. A successor entity, Behold.ai Global Technologies Limited, subsequently acquired the intellectual property and assets and relaunched the platform under new leadership.

South Korean medical AI company whose flagship DeepCARS predicts in-hospital cardiac arrest within 24 hours for patients in general wards, using only four routinely collected vital signs drawn from the EMR: blood pressure, heart rate, respiratory rate, and temperature. Distinctive in this index for operating on biosignals rather than images, in a market where Korean medical AI was largely image based, and for the deployment scale it has reached in one national system, more than 48,000 hospital beds across South Korea including 20 tertiary general hospitals. Approved by Korea's MFDS in 2021 and holding US FDA Breakthrough Device Designation since 2023 while pursuing 510(k) clearance.

Deep learning medical imaging analytics company, formerly Zebra Medical Vision, acquired in November 2021 for up to 200 million dollars and now operating as the AI subsidiary of Nasdaq listed medical imaging hardware company Nano-X Imaging. Its distinguishing strategy is opportunistic population health screening: mining CT scans already acquired for unrelated clinical reasons to surface undiagnosed conditions, with cleared products covering vertebral compression fractures and low bone density associated with osteoporosis via HealthOST, coronary artery calcium quantification for cardiovascular risk, plus pneumothorax and brain bleed detection. Reported ten FDA clearances across the portfolio. Sold to hospitals, HMOs, integrated delivery networks, pharmaceutical companies, and insurers.

Paris based imaging AI company whose Rayvolve suite reads plain radiographs, with AZtrauma covering fractures, joint effusions, and dislocations and AZchest covering lung nodule detection plus pneumothorax and pleural effusion triage. Holds FDA clearances for adult fracture detection since 2022 and pediatric fracture detection since 2024, plus two further chest clearances, and the company states AZtrauma is the only FDA cleared solution in the United States spanning fractures, joint effusions, and dislocations together. Validation is unusually external for a company of its size, drawing on studies run by University Hospitals in Cleveland, SimonMed Imaging, and a 2026 multi-site analysis of 258,373 X-rays across 100 clinical centers in 26 countries.

Ohio based thoracic imaging company whose ClearRead suite applies deep learning vessel suppression to chest CT and radiography for lung nodule detection, cleared by FDA in 2016 as the first device supporting concurrent reading, where the AI output is viewed alongside the initial read rather than after it. The portfolio has since extended to coronary artery calcium scoring on ungated non contrast scans, enabling opportunistic cardiac screening from scans acquired for other reasons. An earlier generation of computer aided detection than most vendors indexed here, with a correspondingly narrow scope, but with a documented multi reader clearance study and a long deployment record.

Radiology AI company operating as a joint venture between an Australian healthcare technology firm and a large radiology provider, offering comprehensive detection across chest X-ray and non contrast head CT. Its enterprise products detect up to 124 findings on chest radiographs and up to 130 findings on head CT, alongside a separately cleared triage and notification portfolio covering ten time critical findings in the United States. Notable for holding Medicare New Technology Add on Payment status and for processing a substantial share of chest X-rays in England.

Gleamer SAS is a Paris based radiology AI company, acquired by RadNet, Inc. on 2 March 2026 in an all cash transaction valued at up to 230 million euros, comprising approximately 215 million at closing plus a 15 million contingent milestone, and integrated into RadNet's digital health subsidiary [[deephealth]]. Ownership, the contracting entity and the hosting arrangement therefore now sit under a United States imaging services operator, which is a change a buyer should confirm before signing rather than discover afterwards. BoneView, its principal product, detects fractures, effusions, dislocations, and bone lesions on radiographs, cleared by FDA for both adult and pediatric use and reported as the only AI fracture detection software holding both. The portfolio has expanded into chest X-ray, pediatric bone age, and musculoskeletal measurement, positioned as semi automated pre diagnosis producing structured reports. Distributed directly and through multiple imaging platform partners including other vendors indexed here. At acquisition the company reported more than 130 staff and more than 700 customer contracts across 44 countries, with annual recurring revenue compounding at more than 90 percent a year from 2022 through 2025 and expected to reach roughly 30 million dollars in 2026. It retains a separate record here rather than being folded into the parent because it continues to carry its own brand, product line, clearances and customer base. The acquirer has stated it intends to deploy these tools across its own network of more than 400 imaging centres, where radiography accounts for close to a quarter of volume.

Cardiovascular imaging company applying AI to coronary CT angiography to quantify and characterize atherosclerotic plaque, producing comprehensive phenotyping of coronary artery disease from non invasive imaging. Distinguished in this index by reimbursement achievement rather than clearance alone: its technology underpins Category I CPT codes, multiple Medicare Administrative Contractor coverage determinations, and commercial payer policies reported to span more than 86 million covered lives. Evidence base includes the multicenter international CONFIRM2 registry.

Cancer focused imaging AI company, publicly listed in Korea, whose INSIGHT suite covers chest X-ray, mammography, and digital breast tomosynthesis. Distinguished less by product breadth than by evidence depth: more than 100 peer reviewed publications spanning The Lancet Digital Health, JAMA Oncology, and Radiology, including third party head to head validations against competing commercial products. Lunit completed its acquisition of Volpara Health Technologies in May 2024. Volpara, founded in 2009 and based in Wellington, New Zealand, made breast screening software covering breast density assessment, imaging quality management and personalised screening, and brought an installed base reported at more than 2,000 United States medical sites. The Volpara name was retired in November 2025, when the business was folded into the Lunit brand and renamed Lunit International, taking commercial operations across the United States, Oceania, Europe and Asia while the Seoul headquarters retained research and new product development. A buyer searching for Volpara or Volpara Health is searching for this vendor. Reported reach after the integration is more than 10,000 healthcare providers across more than 65 countries, with distribution partly through imaging hardware OEM relationships.

AI care coordination platform that pairs disease detection on imaging with automated mobilization of the treating team, originating in stroke where its LVO product received the first ever FDA De Novo authorization for computer aided triage and notification software. The platform now spans more than 50 FDA cleared algorithms across CT, ECG, and echocardiography covering stroke, intracranial hemorrhage, pulmonary embolism, and aortic disease, deployed at over 1,700 hospitals. Its distinguishing mechanism is the mobile app that alerts and connects specialists rather than only flagging a scan.

Lithuanian radiology AI developer whose ChestLink is the first and only regulatory approved application to perform diagnostic reporting autonomously, holding CE Class IIb certification to identify high confidence normal chest X-rays and issue finalized reports with no radiologist involvement. The broader CXR Suite adds detection across 75 findings and a quality module that re checks AI negative studies, with recent CE certified extensions into chest CT and musculoskeletal X-ray. Acquired by imaging IT company Sectra, completed April 2026.

Global radiology AI company with the broadest chest X-ray regulatory footprint in the market, spanning nine products across X-ray and CT. Its qXR suite detects, localizes, and categorizes findings on plain film chest radiographs including lung nodules, pneumothorax, and pleural effusion, alongside qER neurocritical products covering intracranial hemorrhage, cranial fracture, and midline shift. Notably the only chest X-ray computer aided detection device cleared by the FDA with a Predetermined Change Control Plan, allowing model updates without a new submission.

Deep learning MRI image enhancement company whose SwiftMR platform applies AI after image acquisition to remove noise and blur, enabling up to 50 percent scan time reduction or improved image quality at existing scan times. FDA 510(k) cleared and deliberately vendor neutral, working across scanner manufacturers, field strengths, and body parts, including alongside the scanner makers' own deep learning reconstruction pipelines. Seoul based with US operations, reported at more than 1,700 imaging centers and hospitals across over 40 countries.

Radiologist founded company applying AI to detect peripheral vascular disease from CT imaging. Its VascularAssist Occlusion Triage software received FDA 510(k) clearance in June 2026 as a Class II computer aided triage and notification device, flagging suspected vascular occlusion in lower extremity CT for prioritized radiologist review, with the stated goal of catching limb threatening disease earlier. Clinical performance testing supporting the clearance is reported at 95 percent patient level sensitivity on two dimensional analysis and 94 percent on three dimensional, against a threshold of at least one lesion at 50 percent or greater stenosis. No specificity figure has been published alongside those numbers, which matters for a triage device and is discussed on the governance axis. The corporate position changed in mid 2026 and is worth understanding. The operating business is in Boston, but the listed entity is GuideAI Health Corp., a British Columbia corporation formed through a plan of arrangement completed on 16 June 2026, which filed a final non offering prospectus dated 12 June 2026 in British Columbia, Alberta and Ontario and began trading on Cboe Canada under the symbol GDAI at the end of June 2026. That makes the company a reporting issuer whose public filings sit on SEDAR+ rather than with United States securities regulators, which is where a diligence process should look for risk disclosure the website does not carry. Very early commercial stage, with a stated broader ambition in AI driven vascular care.

Dental AI platform that analyzes radiographs in real time, detecting and quantifying caries, periodontal bone loss, calculus, and other pathologies with visual overlays presented chairside. Founded by Harvard School of Dental Medicine and MIT alumni, the company holds multiple FDA clearances across detection and measurement claims and sells to dental groups and dental insurers, with a separate claims review product used on the payer side.

Imaging AI company focused on prostate cancer and theranostics, built around PSMA PET/CT. Three products cover the pathway: DeepPSMA for automated lesion detection and whole body tumor burden quantification, SelectPSMA for predicting which patients will respond to PSMA targeted radioligand therapy, and TrackPSMA for automated longitudinal treatment response evaluation. Products are positioned for clinical and research decision support and are not FDA cleared.

Two products with different jobs, and the second one is new. The build side is infrastructure for making medical imaging AI rather than a clinical application, which made this company structurally different from every other vendor in this lane. Grace is a multimodal foundation model supporting image to image and text to image learning across X-ray, CT, MRI, and echocardiography, trained on over a petabyte of permission based anonymized imaging studies enriched with corresponding reports across 2D, 3D, and longitudinal series. The HOPPR AI Foundry is the surrounding development environment, combining accelerated computing, curated datasets, foundation models, fine tuning tooling, and traceable development workflows inside a stated HIPAA compliant environment, so developers, radiology PACS vendors, and AI companies can build, evaluate, fine tune, validate, and host imaging models without assembling that infrastructure themselves. NVIDIA open models NV-Reason and NV-Generate became available on the Foundry in March 2026, with NV-Reason generating structured analytical reasoning alongside outputs. The MC Chest Radiography Narrative Model, introduced April 2026, is a vision language model translating chest X-rays into structured descriptive text, shipped with training data traceability records and version locking so developers can reproduce results. Forward Deployed Services pairs HOPPR machine learning staff with customer teams for fine tuning. Presto Agent, launched in 2026 and sold separately, moves the company to the opposite end of the same supply chain. It is a model agnostic workflow integration layer that places AI results into a radiologist's existing reporting environment without changes to the picture archiving system, the reporting platform or the templates, and the practice chooses which models run, whether open source, commercial or custom. Radiologists at named private practices describe it running inside their existing reporting application and inserting measurements they would otherwise dictate. The company states that Foundry and Presto are independent products. The combination is worth understanding before purchase: this company now supplies the base models that others build regulated products on, and separately supplies the path by which other manufacturers' outputs reach the radiologist, while being the manufacturer of neither. That position is discussed on the regulatory, oversight and security axes. Founded 2019, led by Dr Khan Siddiqui; $34.5 million raised including a $31.5 million Series A in June 2025, with Health2047, the American Medical Association venture studio, among investors.

Two businesses under one name, and the balance between them has shifted decisively. The company today describes itself as a healthcare technology and services company powering the acceleration of clinical research, and its main navigation is entirely about that network; the device sits in a footer link. SKOUT is a real time computer aided polyp detection device that received FDA 510(k) clearance in 2022 for adults undergoing colorectal cancer screening or surveillance, applying computer vision to endoscopic video to flag suspicious tissue during the procedure. In a randomized controlled trial published in Gastroenterology, SKOUT demonstrated a 27 percent relative increase in adenomas detected per colonoscopy, and the company states it was evaluated in the largest US based multicenter clinical study completed for a computer aided detection device in this category. The vendor reports the device does not extend total procedure or withdrawal time. Distribution runs through an exclusive partnership with Provation, a GI documentation vendor reporting more than 3,500 customer facilities, so for many buyers the commercial counterparty is the distributor. The larger business now is a multispecialty clinical research site network. Iterative Health owns and operates research sites and embeds research into partner provider organisations, using machine learning for patient pre screening and eligibility identification alongside centralized operations and staffing. The network spans more than one hundred sites across North America, Europe, India and Australia, with more than forty pharmaceutical, biotech, device and contract research organisation partners, and the company reports twice the industry benchmark for site activation speed and three times the enrollment rate in inflammatory bowel disease trials. It closed a 77 million dollar Series C in April 2026 led by Intrepid Growth Partners and GV, having entered obesity and cardiology during 2026, appointed a chief medical officer for hepatology and obesity in February 2026, partnered with GI Alliance, One GI and US Heart and Vascular, and acquired three cardiology research sites in Texas in May 2026. Headquarters in Cambridge, Massachusetts and New York. A buyer should treat these as separate propositions. The device covers one procedure in one specialty. The research network covers four specialties, owns physical sites, and does not involve the device. The compliance questions differ sharply between them and are discussed on the regulatory, stewardship and posture axes.

Clinical AI for imaging triage, delivered through aiOS, an enterprise AI operating system that handles data normalization, continuous performance monitoring, and governance so health systems can run multi condition AI without re architecting infrastructure. In January 2026 the FDA cleared what the company describes as healthcare's first comprehensive AI triage solution, powered by CARE (Clinical AI Reasoning Engine), its self developed foundation model. The clearance brought 11 newly cleared indications together with three previously cleared into a single abdomen CT workflow, 14 in total, and per the FDA reviewed pivotal study the new indications reported a mean sensitivity of 97 percent and mean specificity of 98 percent, with the company reporting roughly an order of magnitude reduction in false alerts against single condition tools. The company reports more than 100 million patient cases analyzed on aiOS. A successor model, CARE2, has been announced, and a multi year collaboration with Amazon Web Services supports the foundation model work. Co founded and led by Elad Walach.

Generative AI for radiology reporting and follow up. Omni Impressions generates a report impression from the radiologist's dictated findings, learning each radiologist's language preferences from their historical reports. Omni Reporting auto fills structured templates from natural dictation, and Omni Unchanged pulls stable findings forward from prior reports, which the vendor reports cuts follow up dictation time by half using up to 90 percent fewer words. Continuity tracks incidental findings through to completed follow up; the vendor reports health systems using it improved follow up exam completion from roughly 30 percent to over 75 percent. Works inside existing PACS, RIS, and EHR systems. Series C reached $68 million including strategic investments from four health systems.

Dental AI for radiographic pathology detection. Not to be confused with Pearl Health, a separate company in value based care intelligence also indexed here. Second Opinion analyzes bitewing, periapical, and panoramic radiographs in real time at chairside, highlighting up to 18 findings per image including caries, bone loss, calculus, periapical radiolucencies, and defective restoration margins. Second Opinion 3D extends the platform to CBCT, making Pearl, per the company, the first dental AI company with FDA clearance for both 2D and 3D radiologic analysis. Additional clearances cover pediatric caries detection from age four and AI segmentation. The panoramic clearance was supported by a standalone performance study and a fully crossed multi reader multi case study. The company reports regulatory clearance in 120 countries and a published study in which operator diagnostic accuracy rose from 82 percent to 98 percent with the tool. Founded 2019 by Ophir Tanz.

Deep learning software that enhances and accelerates medical image acquisition on existing scanners rather than detecting pathology. SubtlePET (FDA cleared 2018) and SubtleMR (2019) denoise and sharpen PET and MRI images, allowing faster and lower dose acquisition; SubtleHD (MR) cleared November 2025 and SubtleHD (PET) cleared May 2026 extend that with newer model architectures, with SubtleHD (PET) supporting all FDA approved radiotracers and, per the vendor, up to 75 percent faster PET imaging. A CT product has clearance pending. The company reports 11 FDA clearances, more than 25 peer reviewed publications, and deployment on more than 1,300 scanners, with customers including Mount Sinai, RadNet, and Radiology Partners. Raised $33 million in growth capital anchored by a Series C led by Morgan Stanley Expansion Capital in 2026.