Paige

Computational pathology models are the entire product. The company built its position on deep learning applied to whole slide images for cancer detection support, and has since pursued foundation model scale work in oncology imaging, including a pan cancer detection tool designed to identify both common and rare cancer variants across multiple tissue types. There is no non AI version of the offering.

Positioned as delivering insights to pathologists and oncologists so they arrive more efficiently at precise diagnoses, with the pathologist retaining the diagnostic decision. The prostate product was authorized to support cancer detection rather than to render it, which is the appropriate scoping for software reading slides alongside a specialist. Detailed published description of the review workflow and how confidence is surfaced to the pathologist was not located.

The De Novo authorization pathway required performance evidence reviewed by the FDA and established a new device classification, which is a higher transparency bar than clearing against an existing predicate. Foundation model work in partnership with a major technology company has been described publicly in terms of scale. Model architecture, training data provenance, and current performance figures are not published in detail.

The published instrument does not govern clinical material, and one of its clauses has since operated in a way a buyer should know about. The privacy policy is scoped by its own terms to websites and online services, and the information it enumerates is business and website contact data collected through analytics and marketing tools. It does not describe what happens to the whole slide images a laboratory processes through the platform.

Retention is open ended, stated only as being kept while a business purpose exists or as law requires, with no period and no deletion commitment. Real credit belongs alongside that. The policy allocates responsibility correctly by naming provider customers as primarily responsible for determining use and disclosure, a data protection officer is published with a postal address and telephone number rather than a form, and the training corpus is consistently described as stripped of patient identifiers.

The clause to read carefully states that if the company or substantially all of its assets are acquired, personal data held about customers will be one of the transferred assets. That is boilerplate almost every reader skips, and here it is no longer hypothetical, because the company was acquired in August 2025 while the policy carrying it has an effective date of August 2023 and has not been updated since. Ask for a processing agreement covering slide images, a retention and deletion schedule, a written training position, and confirmation of which entity now holds the data.

The regulatory record is the substance here: the company was first to obtain FDA authorization for an AI based digital pathology product, and subsequently received Breakthrough Device designation for a pan cancer detection tool, a designation intended to expedite review of technologies addressing unmet need rather than a clearance.

Acquisition by a large precision medicine company at a reported 81.25 million dollars is a market validation signal of a particular kind, though a modest figure relative to the company's raised capital, which buyers may read as a comment on standalone commercial traction.

The published instrument does not govern clinical material, and one of its clauses has since operated in a way a buyer should know about. Paige's privacy policy is scoped by its own terms to the company's websites and online services, and the personal information it enumerates is business and website contact data: name, address, email, telephone, IP address and geolocation collected through analytics and marketing tools.

It does not describe what happens to the whole slide images a laboratory processes through the Paige Platform or Alba. Retention is open ended, stated only as being kept for as long as there is a business purpose or as law requires, with no period and no deletion commitment. Storage is stated as being in the United States for the data the policy does cover.

Credited: the policy allocates responsibility correctly by naming provider customers as primarily responsible for determining use and disclosure, a data protection officer is published with a postal address and telephone number, and the training corpus is consistently described as stripped of patient identifiers.

The item a buyer should read carefully is the clause stating that if Paige or substantially all of its assets are acquired, personal data held about customers will be one of the transferred assets. That clause is no longer hypothetical, because the company was acquired in August 2025, and the policy carrying it has an effective date of August 2023 and has not been updated since.

Ask for a data processing agreement covering slide images, a retention and deletion schedule, a written statement on whether customer material informs model development, and confirmation of which entity now holds the data.

The relationship is described correctly and the regime is never named, which is an unusual combination. Paige's privacy policy states that where a patient is treated by one of its provider customers, those provider customers are primarily responsible for determining how Paige uses and discloses the personal information, and directs patients with questions about their health information back to their own provider.

That is a textbook description of the business associate position, in which the covered entity directs the use, and it is a clearer statement of the relationship than several European vendors in this lane manage. What is missing is everything that would let a buyer act on it. HIPAA is not named anywhere. No business associate agreement is referenced, offered or described. Neither the Security Rule nor the Privacy Rule is mentioned.

There is no notice of privacy practices, which is appropriate given Paige is not a covered entity, but nothing substitutes for it either. This matters more here than for the European vendors in this lane, because there is no jurisdictional explanation available: Paige is a New York company whose prostate application holds FDA authorisation and is deployed in United States pathology laboratories, so HIPAA plainly applies to the arrangement.

A buyer should ask for the business associate agreement, and specifically should establish which entity is the counterparty following the company's August 2025 acquisition, since the published policy predates it.

A real certification, disclosed in the least discoverable place available. Paige maintains no security page and no trust centre, and its footer carries only terms and conditions and a privacy policy. Inside that privacy policy, in a single paragraph, the company states that its security measures are evaluated by a qualified third party annually as part of maintaining its ISO 27001 certification, and that any third party receiving personal information must contractually commit to substantially the same level of security.

That downstream obligation is the standard this index looks for and most vendors omit. An annually assessed information security management system is genuine assurance and lifts this above the several vendors in this lane holding nothing at all. What holds it at B is everything around the certification rather than the certification itself.

No certificate is published, no scope or expiry is given, no SOC 2 or HITRUST was located, and there is no penetration testing statement, subprocessor list or vulnerability reporting route. Quality certifications from a British certification body and a medical device single audit programme are reported by third party sources but were not confirmed on the company's own material, and in any case those are quality standards rather than information security ones.

Two currency problems a buyer should raise. The privacy policy carrying this disclosure is dated August 2023, so the certification cannot be assumed current. And Paige was acquired in August 2025, after that date, so the buyer should establish whose information security programme now applies to the contract they are signing and ask for the current certificate to confirm it.

Historically the most significant regulatory milestone in computational pathology. The company obtained the first FDA authorization for an AI based digital pathology product via the De Novo pathway for its prostate cancer detection software, creating the device classification that later pathology AI products clear into. It subsequently received Breakthrough Device designation for a pan cancer detection tool spanning common and rare variants across tissue types. Buyers should note the distinction the company itself observes: Breakthrough designation expedites review and is not authorization, so the pan cancer product should not be treated as cleared.

Paige takes a route to accountability that nothing else in this index has taken, and it is a strong one, but it is not a bias disclosure. In August 2024 the company open sourced its foundation models Virchow and PRISM, and it names its model family publicly, including Virchow, Virchow2, Virchow2G, Virchow2G-Mini and PRISM. Releasing weights matters more than it may appear.

This index has recorded that closedness converts an ordinary technical limitation into a governance failure, because no outside party can characterise where a model degrades and so a buyer cannot audit the boundary. Open weights invert exactly that: any independent researcher can probe these models for failure modes without the company's permission or cooperation. That is external auditability rather than a promise of it.

The corpus is also described more specifically than is usual, with millions of slides drawn from more than a thousand institutions through the Memorial Sloan Kettering relationship, and characterised in the acquiring company's materials as spanning 45 countries and diverse genders, races, ethnicities and regions.

Two limits keep this at B. Corpus composition is not performance, and no breakdown of how any model performs across patient groups was located, which is the same line this index draws for every vendor. And the demographic characterisation comes from the acquisition announcement rather than from Paige's own publications, so it should be confirmed before being relied on.

The company does claim its applications generalise across laboratories regardless of pre analytical and staining variation without additional tuning, which is a strong robustness claim and one that open weights would allow an independent party to test. Publishing subgroup performance would take this to A.

The authorisation route carries this and it is the stronger of the two regulatory paths available. Rather than clearing against an existing predicate, the product went through the pathway that creates a new device classification, which required performance evidence reviewed by the agency and produced published special controls governing every subsequent device of that type.

That gives a buyer an externally set standard to read alongside vendor material, and it brings the reporting, complaint handling and correction obligations that exist whether or not the company advertises them, so a laboratory has a route that does not depend on the vendor granting one. Foundation model work in partnership with a major technology company has also been described publicly in terms of scale. Held below the top grade because the disclosure stops at the pathway.

Model architecture, training data provenance and current performance figures are not published in detail, so a buyer knows the product met a regulatory bar at a point in time and cannot establish how it performs on their own case mix now. No warranty, indemnity or remediation commitment was located. The distinction matters in pathology more than most fields, because scanner, stain and preparation differences between laboratories are exactly what degrades a model that passed elsewhere. Ask for current performance by tissue type on your own scanner and protocol, and for the special controls document.

Strong inside the pathology stack, unaddressed beyond it. Paige ships its own viewing and case management layer rather than depending on someone else's: FullFocus is an FDA cleared whole slide image viewer, and FullFolio is an image management system for case organisation and review, together forming an end to end environment the company offers as a turnkey option.

Its applications are also stated to run within partner digital pathology platforms as well as Paige's own, so a laboratory that has already chosen an image management system is not required to replace it, and the platform is listed on a major cloud marketplace, which gives a procurement route as well as a technical one. That combination of an owned viewer, an owned image management system and availability inside third party platforms is the broadest interoperability position in this lane.

What is absent is the connection outward. No laboratory information system is named, no electronic health record integration is described, and no interface standard such as HL7 or FHIR appears in retrieved materials. Paige Prostate Detect is also stated to require use with specific scanners and viewers, so compatibility is bounded rather than universal and a buyer should confirm their own scanner is in scope. Ask which laboratory information systems results can be returned to, whether reporting is by interface or manual transcription, and for the current list of validated scanners and partner platforms.

Two routes are offered and the company describes the work of adopting either, which is more than most of this lane does. A laboratory can take the applications inside a partner digital pathology platform it already runs, or take the turnkey option and adopt the Paige Platform with its own viewer and image management system.

The company states that it aims for rapid deployment with minimal disruption and describes an onboarding process running from integration through training to support, which is an implementation commitment rather than a compatibility claim. The platform is also distributed through a major cloud marketplace, and the infrastructure is Microsoft Azure, confirmed by the acquiring company assuming Paige's outstanding Azure commitment as part of the 2025 transaction.

Naming the underlying cloud is something most of this lane avoids. Held at B on data residency, which is the half of this axis that is not answered. The only residency statement located covers website data and places it in the United States. Nothing states where slide images are processed or stored, whether any regional options exist for laboratories outside the United States, or how the arrangement changed following the acquisition. For a product distributed through a cloud marketplace and running on a named hyperscaler, those are answerable questions and a buyer should get them in writing.

No pricing of any kind is published. There is no list price, no tier structure, no per slide or per case rate, no subscription band and no worked example, for either the diagnostic applications or the platform. Evaluation runs through a request a trial route and commercial terms through contact with the company, so a laboratory cannot estimate cost or compare against the molecular and gene expression testing this category positions itself against without entering a sales process.

Two things are credited and neither is price. A trial pathway is offered and stated plainly, which is more than several enterprise vendors in this index provide. And distribution through a major cloud marketplace gives a procurement route that can sometimes carry published rates and standard terms, which is worth checking directly.

The gap is more visible here than for a smaller vendor because the products are positioned partly on efficiency and workload relief, and a claim to save pathologist time invites a cost comparison the buyer cannot perform. Following the August 2025 acquisition a buyer should also establish which entity contracts and bills, and whether the applications are now sold standalone or bundled within a wider oncology portfolio, since bundling changes what any quoted figure actually covers.

Oncology pathology, with authorized depth in prostate cancer detection and stated ambition toward pan cancer coverage across common and rare variants in multiple tissue types, the latter designated but not cleared. Buyers are clinical pathology laboratories and, through the parent, precision oncology programs. This has no provider workflow or hospital clinical surface.