Stratipath

A single deep learning model is the product. Stratipath Breast reads routine H&E slides of resected breast tumors and returns a low or high risk group, with no viewer, laboratory or assay of the company's own; it runs through partners' image management systems or the company's portal. The product exists only because the model extracts prognostic signal from images already produced in standard care, positioned as an alternative to gene expression testing.

The artificial intelligence is the product. Remove the model and there is nothing left to sell.

Produces a risk stratification rather than a diagnosis, which places it alongside the clinician rather than in the diagnostic path. The output is a low or high risk group assignment informing treatment intensity decisions, so the oversight question is less about diagnostic error and more about how much weight a clinician places on an algorithmic prognosis when deciding on chemotherapy. The company positions it as decision support, and the risk grouping format keeps interpretation with the clinical team.

The approach and development data are published in the research literature; version and update practice are not. The company's evidence rests on Karolinska Institutet papers (Annals of Oncology for the underlying method, Breast Cancer Research for the validation of the CE marked product) that describe deep learning whole slide image analysis of routine H&E slides for prognostic stratification, name the Swedish cohorts used, and state the endpoints. The validation was run on the CE marked product, which ties the published results to what is sold. No model architecture detail on the product pages, version history or update practice is published.

The approach named without the version and update discipline behind it.

An in house model on the company's own service, with the host and route partners' terms unnamed. Stratipath builds its own deep learning model, developed from Karolinska Institutet research, so no outside model provider sits between a slide and a risk profile. The service runs as a subscription, either through the company's customer portal or through integrations with Proscia's Concentriq Dx, Sectra's digital pathology system and Roche's navify Digital Pathology, and PathAI's AISight menu also lists it. On those routes images pass through the host platform on their way to Stratipath. No hosting provider, region or subprocessor is named, and no retention or training term governs customer images once they arrive.

A structurally short chain at the model layer, with the host and every other party unnamed.

Peer reviewed validation of the marketed product in two independent cohorts, with a head to head against a commercial assay. The company's scientific evidence page cites a validation study published in Breast Cancer Research and conducted by Karolinska Institutet, the company's academic origin, testing the CE marked product's prognostic performance in a retrospective cohort of over 2,700 primary breast cancer patients from two independent hospitals. A comparison with the Prosigna gene expression assay in clinically intermediate hormone receptor positive tumors found Stratipath assigning 49.6 percent low and 50.4 percent high risk against Prosigna's 32.5, 47.0 and 20.5 percent low, intermediate and high, and the approach rests on earlier Karolinska research in Annals of Oncology. For a prognostic claim, validation in independent cohorts with long follow up is the design the claim requires, and the method is readable in the papers.

All published cohorts are Swedish, the studies are retrospective, and company founders come from the same institution, so replication by an unconnected group outside the Nordics remains the open question. Routine use is reported at several Swedish sites, including Unilabs at Capio S:t Göran's Hospital.

Peer reviewed, multi cohort validation with method readable.

The provenance of the development data is documented; nothing is published about customer images. Stratipath Breast is a subscription service, so customer slide images are sent to and processed by the company. The privacy policy covers personal data of business contacts, suppliers and job applicants processed through the company's channels, and does not address patient images, retention, deletion, de-identification or whether submitted cases may be used in model development. The development and validation data, by contrast, are well accounted for in the company's research: Swedish cohorts from Karolinska Institutet and the SCAN-B study, linked to the national breast cancer quality registry under ethical review approvals named in the papers. That answers where the training data came from, not what happens to a customer's cases, and no data processing agreement or safety engineering for handling a wrong output is published.

Nothing published on how customers' patient data moves through the service.

No position under HIPAA or its European equivalent is published for the service. The product page says Stratipath Breast is regulatory aligned for clinical use in the EU and the UK, and the company has said it is not marketed in the United States, so the absence of a business associate position follows from its market.

On the European side, the company publishes neither a processor position nor a data processing agreement for the patient images the service receives; the only privacy policy covers business contacts, suppliers and job applicants, not patient data. Whichever regime applies, no status is stated and no document reaches the product.

Nothing independent is published about the security of a service that receives patient images. Stratipath Breast is delivered as software as a service by subscription, used through integrations or the company's own customer portal, so slide images leave the laboratory and are processed by the company. The site publishes a privacy policy and a cookie policy only, the privacy policy covers business contacts, suppliers and job applicants, and there is no security, trust or compliance page. No ISO 27001 certificate, SOC 2 report, trust center, penetration testing statement, hosting description or vulnerability reporting route is published. Because the data goes to the vendor by design, assurance over the vendor's own environment is the central question here.

Nothing independent published on security.

A European regulatory position stated, with the United States outside the market. Stratipath Breast is marked CE-IVD on the company's pages and described as regulatory aligned for clinical use in the EU and the UK; in August 2026 it received CE marking under the IVDR (EU 2017/746), replacing the earlier IVD directive route. The intended use is prognostic risk profiling of invasive early-stage breast cancer from H&E images of resected tumors, not metastatic recurrences. The company has said the product is not marketed in the United States, and FDA's database lists no clearance for Stratipath. The company calls it the first AI based prognostic risk profiling tool in clinical use, a claim attributed to it.

A non US regime named precisely, with the marketing matching it; no US clearance.

Performance is validated across two hospital cohorts, and nothing is published on performance across patient groups. The Breast Cancer Research validation used two independent Swedish cohorts, which tests the site level generalization that most often breaks prognostic models, and the company describes its data as spanning pre- and postmenopausal patients.

No results by patient group, scanner or staining protocol are presented on the company's pages, no governance framework or monitoring commitment is published, and all published cohorts come from Swedish hospitals, so behavior in more diverse populations is undocumented. For a prognostic score that gap carries weight, since a risk group that behaves differently across populations translates directly into differences in treatment intensity.

A clinician weighs the risk profile, and nothing stands behind it. Stratipath Breast assigns early-stage invasive breast cancers to a low or high risk group to support treatment decisions, and the company presents it as support for physicians, an alternative to gene expression testing rather than a decision on its own. For a prognostic score the error that matters is calibration: predicted risk against observed recurrences in a stated cohort, with the event counts. The company's validation (Breast Cancer Research, over 2,700 patients) is described in terms of prognostic performance and risk group separation, and its Prosigna comparison in terms of agreement between classifications; neither is presented on the company's pages as a calibration figure. No warranty, indemnity or remediation commitment attaches to a result.

Clinician review of a decision support output, with nothing attached to it.

Named image management integrations, with direction undocumented. The product page names integrations with Proscia's Concentriq Dx, Sectra's digital pathology system and Roche's navify Digital Pathology, and PathAI's AISight algorithm menu lists Stratipath Breast; the company also announced a collaboration with Roche to expand access. Laboratories not on those systems use the customer portal. No integration documentation or standard is published, and whether results reach the laboratory information system or the pathology report automatically is not stated. For a pathology product the image management system and LIS, not the patient chart, are the correct surfaces.

Named systems with read access and results returned through them.

A single hosted service with its location unstated. Stratipath Breast is provided as software as a service on a subscription, used either through integration with a laboratory's digital pathology system (Proscia Concentriq Dx, Sectra, Roche navify Digital Pathology) or standalone through the company's customer portal, which means patient images leave the laboratory. The company markets the product for clinical use in the EU and the UK, and names routine use at several Swedish sites, including Unilabs at Capio S:t Göran's Hospital. No hosting provider, region, residency commitment or tenant isolation is published, and nothing states where analysis is performed.

A single hosted option with location implied rather than committed.

No price is published, and the unit is stated. The product page says Stratipath Breast is provided as software as a service by subscription, and the company positions it as a faster and cheaper alternative to gene expression tests, returning a risk profile within the hour instead of weeks. A buyer can therefore frame the cost against a known comparator, the molecular assay it replaces, but no subscription fee, per case rate or tier is published.

The sales posture and unit of charge are discoverable and no price is published.

Deliberately narrow and clearly bounded, with validation behind the indication. Stratipath Breast is for prognostic risk profiling of invasive early-stage breast cancer from H&E images of resected tumors, in pre- and postmenopausal patients, and is not to be used on metastatic recurrences.

It is validated in over 2,700 patients from two independent Swedish cohorts (Karolinska Institutet, published in Breast Cancer Research) and compared with the Prosigna gene expression assay in clinically intermediate tumors, and it is in routine use at several Swedish sites, including Unilabs at Capio S:t Göran's Hospital. The training institutions are named in the research; the company lists no validated scanners or staining protocols, and all published cohorts are Swedish.