Qritive

Models and platform are both first party here, which places this between the pure algorithm vendors and the pure workflow vendors in this lane.

The models are substantial and clinical. The QAi family spans prostate, colon, breast, lymph node and gastric cancer, quantifies immunohistochemistry markers and detects lymph node metastasis. On prostate the module detects malignant glands, identifies tumour architecture, grades to International Society of Urological Pathology criteria, characterises Gleason patterns and quantifies tumour burden. Grading to a published international standard is diagnostic work rather than workflow assistance, and the breadth across five cancer types is wider than most specialists attempt.

The platform is also genuinely a product. Pantheon handles case management, slide viewing, structured reporting, telepathology and remote consultation, supports all major whole slide image formats and integrates with hospital systems. A laboratory could adopt Pantheon as an image management system and use no models at all, and it holds its own medical device registration separately from the modules.

What distinguishes the position is that the company sells its models into competitors' platforms as well as its own, with the prostate module running on Roche's digital pathology platform and integrated with Corista's system. A vendor whose algorithms are good enough that rival platforms carry them has demonstrated the models stand alone commercially.

Graded B rather than A because roughly half of what a customer buys is infrastructure, and below Deep Bio on the neighbouring record, where nothing exists but the models.

The assistive boundary is stated clearly and, unusually, it is stated by the customer as well as by the vendor.

The vendor framing is consistent across years of material: the platform assists pathologists in image analysis, the modules support interpretation, and the technology equips pathologists rather than replacing them. That alone would be ordinary.

What is not ordinary is the customer restating it. In the June 2026 reference laboratory deployment, the reporting states plainly that final clinical decisions remain with pathologists, and the laboratory's own diagnostics chief executive is quoted saying artificial intelligence is intended to support pathologists rather than replace them, adding that grading accuracy in prostate cancer directly influences treatment pathways. A buying institution publicly articulating both the boundary and the reason it matters is stronger evidence that the boundary is real in practice than any vendor statement.

The output design supports oversight. The modules segment and classify regions, flag suspicious areas, and report tumour size and percentage per slide or region of interest, so a pathologist sees where the model looked and what it measured rather than receiving a bare grade. Structured reporting in the platform carries that through to the signed case.

What holds this at B is that nothing published describes escalation or disagreement handling, what happens when the model and the pathologist differ on grade, or whether disagreements are recorded. Grading is precisely where a second opinion can anchor a reader, and no anchoring safeguard is described.

Graded B.

Function is described well, performance is described not at all, and the gap between the two is the finding.

The functional detail is good. The prostate module identifies prostatic adenocarcinoma, segments and classifies benign and malignant areas, works from both whole slide images and selected regions of interest, and reports tumour size and percentage per slide or region. The more recent description adds detection of malignant glands, identification of tumour architecture, grading to International Society of Urological Pathology criteria, characterisation of Gleason patterns, quantification of tumour burden and flagging of suspicious regions. Naming the international grading standard rather than a proprietary score is important, because it means output can be checked against a framework pathologists already use. Deep learning is identified as the method and the platform is described concretely as vendor agnostic, cloud ready and supporting all major whole slide image formats.

What is absent is every number. No sensitivity, specificity, concordance or accuracy figure was located for any of the seven module capabilities, no processing time is stated, no training data is described, no evaluation method is published, and no versioning or update cadence is given. The neighbouring Deep Bio record publishes sensitivity, specificity, two concordance figures and the reference standard construction; this one publishes a capability list.

For a module that grades cancer severity, the absence of any published accuracy figure is the single largest gap on this record.

Graded C.

Distribution partners are fully named, the infrastructure partners are named, and the models' own provenance is undisclosed.

What is traceable is where the intelligence goes and what it runs on. Roche's digital pathology platform and Corista's DP3 are both named as carrying the modules, so a customer can identify exactly which model is running inside which platform. Infrastructure and technology relationships are named openly, including major cloud and hardware providers and a national hospital, which tells a buyer what the platform is built on. Founder backgrounds are documented in a way that is itself provenance of a sort, with the chief executive's doctorate in mechanobiology involving computer vision for identifying cancer cells and the chief technology officer's in natural language processing, which indicates the models are in house work built on that foundation.

What is missing is everything upstream of the models. No training corpus is described for any of the seven capabilities, no source institutions are named, no acknowledgement of any pretrained component or open framework, and no bill of materials. For a company shipping across five cancer types the absence of any account of what any model learned from is the substantive gap.

ISO 13485 certification implies design control documentation exists internally, including assessment of software of unknown provenance, which is a regulatory artefact rather than a published one.

Graded C.

Deployment evidence is strong, geographically distinctive and independently corroborated, while published performance evidence is thin.

The deployments are the substance and they are named institutions rather than logos. Three Indian organisations adopted the platform and modules together: a large diagnostics chain, a dedicated cancer institute and a specialist diagnostics company, with the chain executive quoted specifically on prostate cancer detection. In June 2026 M42's National Reference Laboratory integrated the prostate module at Cleveland Clinic Abu Dhabi, reported by a third party rather than only by the vendor, with the laboratory's own diagnostics chief executive commenting that grading accuracy in prostate cancer directly influences treatment pathways. Partnerships place the modules inside Roche's platform and Corista's system, and a national industry survey independently records the Roche deployment.

That spread across Singapore, India and the Gulf is unusual and it matters, because it demonstrates the models being adopted across populations and laboratory practices that differ considerably from one another.

What is missing is measurement. No sensitivity, specificity or concordance figures were located for any QAi module, no peer reviewed validation study was found, and no independent evaluation of performance exists in anything examined. The neighbouring Deep Bio record publishes concordance figures against a three pathologist reference standard and an external validation study in a discipline journal; this one publishes adoption instead.

Graded B on deployment breadth and independent reporting, held below the top grade by the absence of any published performance number.

Nothing published addresses training data, retention or secondary use, and the deployment pattern makes the training question the interesting one.

No training corpus is described for any of the seven module capabilities. No size, no source institutions, no consent or research authorisation basis, no de identification standard and no statement about whether customer images inform continued development. For a company shipping models across five cancer types, the absence of any account of what they were built on is the central gap on this record.

The geographic spread makes it matter more than usual and cuts both ways. Deployments across Singapore, India and the Gulf mean the models meet tissue prepared under materially different laboratory practices, and if development data came predominantly from one region the others are effectively operating on generalisation the vendor has not described. Conversely, if the company has assembled genuinely multi regional training data that would be a real strength and a competitive one, and it is not claimed.

The platform position compounds the retention question. Pantheon is an image management system, so a customer's entire whole slide image archive may sit inside vendor supplied infrastructure described as cloud ready, and nothing states where it lives, how long it is kept or what happens at contract end.

The modules running inside Roche's and Corista's platforms create a third undescribed path, where images processed by a Qritive model sit within a third party environment under terms nobody publishes.

Graded D on the absence.

No published position was located, and this record has a jurisdictional profile that makes the gap unusually consequential.

The company is Singaporean, with operations in India and the United States, and live deployments in Singapore, India and the United Arab Emirates. Each of those is a distinct data protection regime: Singapore's personal data legislation, India's more recent digital personal data protection law, Gulf state health data rules, European data protection law reached through the CE mark, and the United States health privacy statute for any American operation. A platform holding whole slide images with patient and case data across all of them faces cross border questions in several directions at once.

Nothing published addresses any of it. No business associate agreement is offered or described, no protected data handling summary exists, no retention position is stated, no cross border transfer mechanism is described, and no statement addresses where images processed by the modules are held.

The two sided architecture adds a second question. When the prostate module runs inside Roche's or Corista's platform, patient images are processed by Qritive's model within another vendor's environment, and nothing describes what that arrangement means for data handling or which party is the processor.

One credential is relevant and is not a privacy posture. ISO 13485 certification covers design, development, manufacture, distribution and installation of software medical devices, which is quality management rather than data protection, and Singaporean device registration entails regulatory scrutiny in one jurisdiction.

Graded D.

One substantive certification, aimed at product quality rather than information security, and nothing else.

The certification is ISO 13485, covering design, development, manufacture, distribution and installation of software medical devices, and the company describes it as held internationally. That is a genuine external audit of how the product is built, changed and maintained, including design controls and risk management, and for a company shipping diagnostic software across three continents it is the right foundation. Medical device registration in Singapore and CE marking add regulatory scrutiny of the software lifecycle in two more regimes.

What is absent is any information security assurance directed at a customer's security team. No trust centre, no service organisation control report, no information security management certification, no penetration testing statement, no vulnerability disclosure policy, no subprocessor list and no incident notification commitment were located.

That gap is more noticeable than usual because the platform is an image management system, meaning it holds a laboratory's whole slide archive and case data rather than analysing images passed to it, and because named partnerships with large cloud and hardware providers indicate infrastructure sophisticated enough to have a documented security posture behind it.

The Gulf reference laboratory deployment implies a demanding security review was passed, since a laboratory operating under an international hospital brand does not integrate a diagnostic module without one. As elsewhere on this record, the assurance exists privately for one customer and is published for none.

Graded C.

Genuine multi jurisdiction registration at a low risk class, with no clearance located for the diagnostic modules themselves.

What is documented is the platform. Pantheon is registered with Singapore's Health Sciences Authority as a class A medical device and is CE marked, described in one account as CE marked for in vitro diagnostic use. The company also holds ISO 13485 certification covering design, development, manufacture, distribution and installation of software medical devices, which is the quality management standard underpinning device manufacture and is a substantive external audit.

The distinction that sets the grade is class and scope. Class A is the lowest risk tier in the Singaporean system, appropriate for an image management and viewing platform, and it is a materially different credential from the class 3 in vitro diagnostic approvals held by Deep Bio on the neighbouring record for algorithms that grade cancer severity. Nothing located establishes an equivalent approval for the QAi modules as diagnostic devices in their own right, and grading to International Society of Urological Pathology criteria is the kind of output that would ordinarily attract a higher classification.

No United States clearance was located despite the company having United States operations.

The practical consequence for a buyer is that regulatory assurance here covers the platform holding and displaying images, and the status of the algorithms interpreting them is not established in anything published. That is the question to resolve directly.

Graded C.

Nothing was located. No model card, no training data description, no performance figures for any module, no subgroup analysis and no bias statement.

The exposure is unusually concrete here because the deployment footprint spans populations that differ in ways known to affect these specific models.

Prostate cancer incidence, presentation and Gleason pattern distribution vary by ancestry, and this module is deployed across Singaporean, Indian and Gulf populations. Gastric cancer, which the module family also covers, has incidence patterns concentrated in East Asia and different presentation from Western cohorts. Laboratory practice differs too: staining protocols, section thickness and scanner types vary between a large Indian private diagnostics chain, a Gulf reference laboratory and a Singaporean hospital. A model developed predominantly in one setting and deployed across all three is being asked to generalise across exactly the variables that matter, and nothing published examines whether it does.

That spread could be a strength rather than a risk if the training data reflects it, which would be a genuine competitive advantage over vendors built on Western cohorts, and the company does not claim it.

The grading task carries the additional problem that the reference standard is itself variable, since inter observer disagreement in Gleason grading is well documented, and no breakdown by grade group is published for a model whose output changes treatment at grade boundaries.

Graded D on the absence, with the multi population deployment recorded because it raises the stakes rather than lowering them.

Nothing published addresses responsibility for an automated outcome, and the stakes are stated on this record more explicitly than on most, by the customer rather than the vendor.

The M42 reference laboratory deployment carries a quote from its diagnostics chief executive noting that grading accuracy in prostate cancer directly influences treatment pathways. That is exactly right: a Gleason grade determines whether a patient is offered active surveillance or radical treatment, so an error in either direction carries serious consequence. A module that grades to International Society of Urological Pathology criteria is producing an output on which a treatment decision turns, and no published position states who bears responsibility when it is wrong.

The two sided architecture creates a division that is harder here than on most records. A laboratory running the prostate module inside Roche's platform has a contract with Roche, receives a result generated by Qritive's model, and may have no direct relationship with the party that built it. Nothing states what either party warrants, or where a complaint about a grading error is directed.

The regulatory position offers less shelter than it might. Class A registration covers the platform, and no equivalent approval was located for the modules, so the assurance a device classification would provide for the algorithm itself is not established.

One genuine mitigation exists: the customer statement that final clinical decisions remain with pathologists is a public allocation of responsibility, made by a buying institution.

No indemnity, limitation, performance warranty or accuracy commitment was located. Graded D.

Interoperability is the architectural commitment here and it is demonstrated in both directions, which very few vendors in this index can claim.

Inward, Pantheon is vendor agnostic and supports all major whole slide image formats, so a laboratory can adopt it without replacing scanners, and the platform is described as integrating with hospital systems and supporting telepathology and remote consultation. Scanner neutrality in digital pathology is a substantive commitment rather than a courtesy, because proprietary format lock in is the defining constraint of the field.

Outward is the unusual part. The company's own modules run inside competitors' platforms: prostate grading deployed on Roche's digital pathology platform, and the module family integrated with Corista's DP3, which is described as standards compliant for medical imaging and able to work with any whole slide scanning device and laboratory information system. A company that sells an image management system and simultaneously makes its algorithms available inside rival image management systems has accepted that customers may choose a competitor's platform and still want its intelligence. That is interoperability as a commercial position rather than a technical feature.

The combined integration is described as enabling pre reviewed analysis of images from any scanner and any laboratory information system, and structured reporting carries results into the signed case rather than leaving them in a viewer.

Graded A: neutrality claimed, and evidenced by the company's own willingness to run inside competitors' products.

One architectural fact is disclosed and every jurisdictional specific is not, on a record where jurisdiction is more complicated than anywhere else in this lane.

The disclosed fact is that Pantheon is cloud ready and scalable across diverse clinical settings, and the company is described as working with major cloud and hardware providers. Cloud ready implies the platform can also be deployed otherwise, which for a laboratory in a jurisdiction with data localisation requirements is the important flexibility, and the phrasing suggests both options exist without confirming either.

Everything specific is absent. No hosting region, no residency option, no subprocessor list, no retention position, no export or contract end terms and no availability commitment were located.

The jurisdictional picture makes this the sharpest version of the question in the pathology lane. The vendor is Singaporean with operations in India and the United States, and live deployments in Singapore, India and the United Arab Emirates. India's data protection law and Gulf health data rules both carry localisation expectations for health information, and a Gulf reference laboratory attached to an international hospital brand will have required a specific answer. That answer exists privately for at least one customer and is published for none.

The two sided architecture adds a further path. When the prostate module runs inside Roche's or Corista's platform, execution happens in that vendor's environment under that vendor's residency terms, so a customer's data location depends on which route they took to the same model, and nothing describes either.

Graded C on the strength of the cloud ready disclosure alone.

Nothing is published. No price, no unit of charge, no module licensing structure, no platform cost, no implementation fee, no contract term and no minimum was located in vendor or third party material.

The two sided architecture makes the absence more consequential than a single product would. A buyer can acquire the platform alone, the modules alone through a competitor's platform, or both together, and those are three different commercial relationships with three different cost structures. Nothing indicates whether modules are licensed per cancer type, per case analysed, per site or bundled with the platform, and nothing states whether a laboratory running the prostate module inside Roche's platform pays Roche, Qritive or both.

The module count sharpens it. With five cancer types plus marker quantification and metastasis detection, whether a laboratory buys the family or selects individual modules determines the cost entirely, and a laboratory that starts with prostate has no published basis for estimating what adding colon or breast would cost later.

One structural signal exists without a number attached. The company sells into public and private institutions across Singapore, India and the Gulf, markets with very different price sensitivity, which implies pricing varies by geography and makes a single published figure unlikely. That is an explanation rather than a justification, and a published range or unit basis would still be possible.

Graded D.

The broadest coverage of any analysis vendor in this lane, across cancer types, institution types and geographies simultaneously.

Clinical breadth is the strongest element. The module family covers prostate, colon, breast, lymph node and gastric cancer, plus immunohistochemistry marker quantification and lymph node metastasis detection. Gastric is worth noting on its own: it is a leading cancer across East Asia and it is comparatively neglected by vendors built around Western case mix, so a company serving Asian markets building for it is addressing real regional need rather than following the field.

Institutional coverage spans hospitals, diagnostic centres and pharmaceutical companies, and the named deployments bear that out across a large private diagnostics chain, a dedicated cancer institute, a specialist diagnostics company and a national reference laboratory attached to an international hospital brand.

Geographic coverage is the distinctive part. Singapore for the home market and regulatory anchor, India through three named institutions, the United Arab Emirates through the reference laboratory deployment, Europe through CE marking, and operations in the United States. Very few companies of this size have live clinical deployments across three continents, and doing so means the models have met genuinely different tissue preparation practices and patient populations.

Platform coverage completes it, with support for all major whole slide image formats making the system scanner agnostic and telepathology extending reach to laboratories without on site subspecialty expertise.

Graded A.