Mindpeak

Deep learning models are the whole company. The portfolio is a set of biomarker quantification algorithms operating on H and E, immunohistochemistry, and multiplex immunofluorescence images, with no platform, scanner, or assay of its own. That purity is also a commercial dependency: the algorithms need someone else's workspace to reach a pathologist, which is why they are distributed through PathAI's platform as well as directly.

Assistive quantification with the pathologist interpreting. The distinguishing design question in this category, as sector commentary notes, is whether the pathologist manually selects the region of interest or the system analyzes the whole slide unaided, and Mindpeak ships both patterns, with region of interest products named as such alongside broader analysis. Peer reviewed evaluation frames the tool as improving interobserver agreement rather than replacing the reader. Clear assistive positioning; the human sets the clinical meaning of the score.

The technical claim is specific and checkable: BreastIHC was described at launch as the first deep learning solution able to distinguish tumorous from non tumorous structures at the cellular level in single stained slides, classifying cells into positively stained tumor and unstained tumor populations. The company markets plug and play deployment without per site calibration, which is a meaningful robustness claim.

Training data scale and architecture are not published, and the robustness claim is asserted rather than independently benchmarked, but independent studies have tested the tools across image variance.

This vendor publishes a dedicated image upload policy, which is the instrument its lane peers lack entirely, and reading it matters because it does not protect the uploading laboratory, it grants rights against it. By having images analysed, the uploading party grants the company and its affiliates a non exclusive, royalty free, perpetual right, unrestricted as to territory, time and content, to copy, transmit, reproduce, publicly display, publicly perform, edit and reformat the images, together with the right to sublicense.

The stated purposes include service provision and documentation duties, and also advertisement and product improvement, and the clause names the retraining of artificial intelligence expressly. The burden is then placed on the customer, which must represent and warrant that it owns all rights necessary to grant that licence, over material that belongs to patients rather than to the laboratory.

So the question this index puts to every vendor is resolved here, in the direction least favourable to the buyer. Three aggravating points. The clause is not scoped to demonstration or trial uploads and nothing distinguishes a demonstration image from a clinical one. Nothing addresses de identification, retention or deletion, and slides carry embedded label images by default.

And the page carrying it is marked so that search engines will not index it, so the single document that most affects what happens to uploaded patient tissue cannot be found by searching. Disclosure that cannot be found is not disclosure. Ask whether the clause reaches clinical use and whether it can be struck.

Genuinely independent evidence exists, which is rarer than it should be here. A study by Abele and colleagues assessed the Breast Ki-67 and ER/PR products for quantifying Ki-67, ER, and PR, finding reliability across a wide range of image variance and pathologist confirmation, and concluding the tools can improve interobserver agreement in real world clinical settings.

A separate PubMed indexed concordance study at Aga Khan University Hospital compared automated Ki67 scoring against expert pathologist eyeball assessment across 60 invasive ductal carcinoma cases and reported strong positive concordance. Both are third party rather than vendor run, though sample sizes are modest and endpoints are concordance rather than patient outcome.

Mindpeak publishes a dedicated image upload policy, which is the instrument its European peers in this lane lack entirely, and reading it matters because it does not protect the uploading laboratory, it grants rights against it.

The policy states that by having images analysed, the uploading party grants Mindpeak and its affiliates a non-exclusive, royalty free, perpetual right, unrestricted as to territory, time and content, to copy, transmit, reproduce, publicly display, publicly perform, edit and reformat the images, together with the right to sublicense those rights.

The stated purposes include service provision and documentation duties, and also advertisement and product improvement, and the clause names the retraining of artificial intelligence expressly. It then places the burden on the customer, which must represent and warrant that it owns all rights necessary to grant that licence.

This resolves the question this index puts to every AI vendor, and it resolves it in the direction least favourable to the buyer: customer images are used to retrain models, and may also be publicly displayed and used in advertising. Three further points deserve weight. The clause is not scoped to demonstration or trial uploads, and nothing distinguishes a demonstration image from a clinical one. Nothing in it addresses de identification, retention or deletion.

And the page carrying it is marked noindex, so search engines cannot surface the single document that most affects what happens to uploaded patient tissue. The general terms and conditions were not opened on this pass and may scope the clause further. Note that this grade sits at C for a different reason from its lane peers: Aiforia and Visiopharm publish nothing covering clinical images, which is silence, whereas here an instrument exists and it operates against the uploading laboratory. Ask whether the clause applies to clinical as well as evaluation use, whether images are de identified before any retraining or advertising use, and whether it can be struck from the contract.

No HIPAA position was located. There is no HIPAA notice, no statement of business associate status, and no reference to a business associate agreement or to the Security or Privacy Rules. The footer legal section carries a privacy policy, an image upload policy, general terms and conditions and a carbon reduction plan, and nothing on HIPAA.

That absence carries more weight here than for the other European vendors in this lane, because Mindpeak operates a United States legal entity registered in Cambridge, Massachusetts, and names United States customers including a large integrated health system and a national reference laboratory. This is a company selling into US laboratories with no published HIPAA position at all.

It becomes sharper when read against the company's own image upload policy, which grants a perpetual, sublicensable right to use uploaded images for advertising and for retraining artificial intelligence. A business associate agreement would not ordinarily permit a vendor to use protected health information for advertising, so a US laboratory needs to establish which of two situations applies: the uploaded images are de identified and therefore fall outside HIPAA, or they are protected health information, in which case the upload licence and the vendor's obligations need reconciling. Nothing published states which. Ask for the business associate agreement and for a written statement on de identification before uploading anything.

No information security management system certification was located. No ISO 27001, no SOC 2 of either type, no HITRUST and no trust centre. One real disclosure is credited, and it is more than several peers in this lane offer: a penetration test certification badge is displayed alongside the company's other credentials and carries a date of December 2025, which at least establishes that independent testing was carried out and when.

ISO 13485 quality management certification is held and displayed with a certification body assurance mark, but that is a medical device quality standard rather than an information security one and is not counted here as security. The company also markets audit trails as a built in feature.

What is absent is the assurance layer a laboratory would normally require: no independent attestation of security controls, no published penetration test scope or findings summary, no description of encryption or access control, and no subprocessor list. That last omission matters more than usual here, because the company's image upload policy grants a right to sublicense uploaded images to third parties and no such party is named anywhere. Ask for an ISO 27001 or SOC 2 position and timeline, for the penetration test scope, and for the identity of any party to whom images may be sublicensed.

Eleven CE-IVD marked algorithms is a substantial European portfolio, covering breast HER2, ER, PR, and Ki-67 plus lung PD-L1, and the company reached CE-IVD for primary diagnosis with BreastIHC as an early mover. No FDA clearance was located, and the company is explicit that its prostate and other tumor type assays are research use only rather than clinical. Graded C on the established basis: strong non US authorization, no US clinical clearance, with clear disclosure of which products are research use only.

The company addresses bias directly, which is more than most of this lane manages, but it addresses it at the level of assertion. A published answer states that Mindpeak prioritises diverse data curation, cross site validation and continuous monitoring to support responsible and generalisable deployment.

Those are the right three practices to name, and cross site validation is particularly relevant in pathology, where staining protocol and scanner variation between laboratories is a genuine source of model failure. But no method, no measurement, no threshold and no result accompanies any of them. There is no performance breakdown by any patient group, no account of what diverse curation meant in practice, and no published output from the monitoring.

Against a portfolio spanning breast, lung, gastric, oesophageal, urothelial, ovarian and prostate indications, that is a large amount of unexamined surface. One connection a buyer should make, because the two statements live in different documents and belong together: the governance answer refers to continuous monitoring, while the company's image upload policy grants a perpetual right to use uploaded customer images for the retraining of artificial intelligence.

Retraining on customer material is a specific and consequential governance practice, and here it is disclosed in a licensing document rather than in the answer about how the company governs its models. Publishing a subgroup and cross site performance breakdown, and describing the retraining process and the controls around it, would move this materially.

The technical claim here is specific enough to be checkable, which is what separates it from the category norm. The product was described at launch as the first deep learning solution able to distinguish tumorous from non tumorous structures at the cellular level in single stained slides, classifying cells into positively stained tumour and unstained tumour populations.

That is a claim about a defined capability on a defined input, and it is the kind a competitor or a reviewer can dispute rather than an atmospheric assertion. Independent studies have also tested the tools across image variance, which is external examination of the property that matters most in this field. The robustness claim is the part to press.

Plug and play deployment without per site calibration is a meaningful assertion, because per site calibration is what most pathology models need in order to survive different scanners, stains and preparation practices, and a product that genuinely does not need it is doing something harder. It is asserted rather than independently benchmarked, and it is exactly the claim a laboratory should test on its own slides before relying on it.

No training data scale, architecture, accuracy figure or warranty, indemnity or remediation commitment was located. Ask for performance across scanner and stain protocols, what happens on slides outside the tested variance, and for the independent study results.

Reach comes through platform distribution rather than direct integration engineering. Since 2025 the CE-IVD portfolio is accessible inside PathAI's AISight platform, which is separately indexed here and itself FDA cleared for primary diagnosis in the US and CE-IVD marked in Europe. For a laboratory already running that platform, adoption is algorithm selection rather than an integration project.

For everyone else, integration depends on which host workspace is in place, and the vendor does not publish a broader connector list. This is the Archy and Pearl pattern: the algorithm reaches the pathologist through someone else's product.

The most specific integration position in this lane, stated in numbers and backed by named institutions. Mindpeak states its technology is in use in more than 40 laboratories globally and describes more than 18 clinical image management system partner integrations, presented as the most comprehensive integration ecosystem in the category.

The proposition is direct connection to existing scanners, laboratory information systems and cloud storage, with no replacement of existing infrastructure and no downtime. Named customer institutions span several countries and both academic and commercial laboratory settings, including university hospitals in Germany, a European laboratory network, a large United States integrated health system and a national United States reference laboratory.

The company operates two legal entities, one in Hamburg and one in Cambridge, Massachusetts. Held at B rather than A on three points. The integration partners are given as a count rather than enumerated, so a buyer cannot check whether their own image management system is among them.

There is no data residency commitment and no statement of where analysis is performed or where uploaded images are stored, which matters more than usual given the breadth of the rights the company's image upload policy takes over those images. And no implementation timeline or resourcing expectation is published, unlike vendors elsewhere in this index that describe the deployment process itself.

No pricing is published. The dual route to market, direct and embedded in PathAI's platform, means commercial terms likely differ by channel, and neither is disclosed. A buyer should establish which route they are purchasing through, since it affects both cost structure and who holds the support and data relationship.

Concentrated on breast cancer immunohistochemistry, the highest volume biomarker workload in European pathology, extending to lung PD-L1 clinically and to prostate and other tumor types for research use only. Depth in a high demand area rather than breadth across pathology, and the clinical versus research use split means actual coverage is narrower than the full product list suggests.