Proscia

A digital pathology platform with AI offered on it, and the company is careful about which parts are which. Concentriq is first an image management, viewing and workflow system for whole slide pathology, which is what high volume laboratories buy, and the cleared clinical edition (Concentriq AP-Dx) is an aid for viewing, managing and interpreting slide images for primary diagnosis. The home page says Proscia's native AI features in Concentriq AP-Dx "do not meet the definition of Artificial Intelligence as defined by Regulation (EU) 2024/1689", that Proscia's AI applications are for research use only, and that more than 230 AI applications, most from other manufacturers, run on the platform under their own makers' labeling. The company describes its own models as domain specific vision, language and multimodal models that recommend stain panels and score PD-L1, HER2 and Ki-67 in the research editions.

The platform's value stands without the models, and the AI is a layer offered on it, much of it by third parties.

AI as a feature layer on a platform that stands without it.

Built around expert verification, with the failure side unpublished. Outputs are described as backed by spatial, contextual and quantitative evidence and built for expert verification; in practice the pathologist reviews AI recommended stain panels and adjusts them inline, and can modify regions of interest on biomarker scoring with results updating in real time. The cleared clinical edition is an aid to the pathologist for viewing and interpreting images, and the company's AI applications are for research use only, so no model output reaches a diagnosis without a pathologist. The company's chief executive has publicly argued that the opportunity is removing administrative burden rather than diagnostic expertise, consistent with how the product is scoped.

The part missing is what happens when a model is wrong: no published error behavior, known failure conditions (scanner, stain, tissue type) or route for reporting a wrong output.

Review structure described with the failure handling missing.

The architecture is described specifically as domain specific vision, language, and multimodal models reasoning across images and data, with images, metadata, and clinical context unified in one layer. Early validation practice was strong and disclosed: its dermatopathology classifier was validated in a multi site study across named academic and commercial laboratories using more than 20,000 patient biopsy slides. Falls short of an A because current model performance figures and validation methodology for the shipping AI features are not published at that same level of detail.

The chain is short for Proscia's own models and open for the platform's other parties. The platform page names Amazon Web Services as the host, and Proscia's own models are its own; customers can run their own models as containerized workloads alongside their data. The platform also hosts more than 230 AI applications, most from other manufacturers, each with its own labeling that Proscia says it does not warrant, so an image sent to a third party application reaches that maker's model under terms the buyer must obtain separately. The platform page commits that customer data is never used to train the models Concentriq runs and that de-identified sharing for research happens only with the customer's express permission.

No subprocessor list is published, and nothing states which third party applications process data inside the customer's environment and which send it elsewhere.

A short chain for the company's own models with the host named, and the third party layer not enumerated.

Named deployments with dated outcome figures, and a multi site validation of an earlier model. The company reports a 68 percent improvement in one day case turnaround at a named laboratory group, 60 percent faster workflow cycles at a contract research organization, a 40 percent reduction in time spent searching for data, and more than $1 million of new revenue at a named pathology group, with no method or baseline published; it cites a Best in KLAS 2026 award and states that all customers interviewed by KLAS would buy Concentriq again, which is satisfaction rather than outcome evidence. Clearance of the clinical edition for primary diagnosis rests on FDA review of the viewer (K230839, K261380). The earlier dermatopathology classifier was validated across two university hospitals and two commercial dermatopathology laboratories on more than 20,000 biopsy slides. Scale is stated: more than 4,000 hospitals and clinics, 12 million cases a year, and 16 of the top 20 pharmaceutical companies.

No published validation covers the current AI applications, which are research use only.

Named deployments with figures and published research short of independent validation.

A categorical commitment on training use and sharing, without a retention schedule. The platform page states that "everything in Concentriq stays within your organization, unless you give express permission to de-identify and share it to advance research", and that "your data is never used to train the models Concentriq runs". That is the categorical form of the commitment rather than one excluding only identified data, and the research sharing exception is published beside it as opt in. Customer data is described as portable through open APIs, customer built models run as containerized workloads alongside the data with intellectual property retained by the customer, and the platform runs on AWS under ISO 27001 and SOC 2 Type II programs.

What is not published is the rest of the lifecycle: no retention period, no deletion position at contract end, no de-identification method for the opt in research sharing, and no subprocessor list. Safety engineering, meaning how a wrong model output is handled as a safety event, is not published.

HIPAA is claimed as alignment, and no business associate position or service privacy document is published. The platform page lists HIPAA alignment and GDPR compliance beside ISO 27001, ISO 13485 and SOC 2 Type II, and states that customer data stays within the customer's organization unless the customer permits de-identified sharing and is never used to train the models Concentriq runs. The privacy policy linked in the footer is the company's website privacy policy. No business associate agreement is named, offered or described, and nothing sets out which obligations Proscia accepts when it hosts diagnostic images for clinical laboratories, which include major United States health systems, a national reference laboratory network and a federal pathology center.

Alignment is not a stated business associate position, and the ISO 27001 and SOC 2 programs speak to security rather than to business associate terms.

Compliance claimed without a business associate position or a service privacy document.

Recognized certifications are named on the company's own pages, without the artifacts. The platform page states that Proscia's security and quality programs are certified under ISO 27001, ISO 13485 and SOC 2 Type II, beside HIPAA alignment and GDPR compliance, and adds two operational commitments: a 99.95 percent uptime service level agreement with regional redundancy, and change control in which every change is classified and validated by risk with a record of how each result was produced. The scale protected is substantial, with more than 8 petabytes of research data and more than 6,000 cases an hour.

No certificate, report, scope statement, expiry date or trust center is published, so the claims cannot be verified without contacting the company; no penetration testing statement or subprocessor list appears. The home page does not repeat the certifications.

The regulatory position is identifiable and matches what is sold. FDA's database lists two 510(k) clearances for Proscia, Inc.: Concentriq Dx (K230839, 8 February 2024, product code PSY) and Concentriq AP-Dx (K261380, 5 August 2026, QKQ), both under 21 CFR 864.3700, whole slide imaging systems. The home page states the intended use: an aid to the pathologist for viewing, managing, reviewing and interpreting digital images of scanned surgical pathology slides prepared from FFPE tissue, for primary diagnosis, and not for frozen sections, cytology or non-FFPE hematopathology specimens; the edition is also Health Canada licensed and CE marked under the EU IVDR. The August 2026 clearance adds a Predetermined Change Control Plan, cloud deployment and the Leica Aperio GT 450 DX scanner.

The AI is positioned consistently with that scope. The research editions (Concentriq AP and LS) are labeled research use only; Proscia's AI applications are research use only; third party applications carry their makers' labeling, which Proscia says it does not warrant; and the company states its native features in AP-Dx do not meet the EU AI Act's definition of artificial intelligence. No AI function is called cleared.

Evaluation practice is described without a governance framework or results by group. The company states that outputs are interpretable and evidence backed, with overlay tools letting pathologists confirm results visually, and that every change is classified and validated by risk with a record of how each result was produced. Its earlier dermatopathology classifier was validated across four laboratories (two university hospitals and two commercial dermatopathology laboratories) on more than 20,000 biopsy slides. No analysis is published of performance across scanner platforms, staining protocols or patient demographics, which are the known sources of drift in computational pathology, although the platform supports more than 45 scanners. The cleared edition's change control concerns the viewer, not a model, and third party AI applications carry their own makers' governance.

Evaluation practice described with no results by scanner, stain or patient group.

Mechanisms let a pathologist check every output, and nothing stands behind it. The pathologist reviews AI recommended stain panels and adjusts them inline, and can modify regions of interest on biomarker scoring with results updating in real time; overlays show the evidence behind each result; and the platform page states that every change is classified and validated by risk with a clear record of how each result was produced, which is an audit trail a laboratory can use to reconstruct how a case was analyzed. The company's AI applications are for research use only, and for third party applications the home page says their regulatory status and intended use are set by their own manufacturers and "not warranted by Proscia".

No warranty, indemnity or remediation commitment attaches to a model's output, and no error rate is published for the shipping models; the multi site dermatopathology validation described an earlier classifier. The 99.95 percent uptime commitment concerns availability, not output.

Review before sign out and result level provenance, with nothing attached to the output.

Integration is with the laboratory information system rather than the chart, which is the correct surface. The company states that AI recommended ancillary test requests flow back into the LIS rather than a separate ticket queue, which is a real workflow closure rather than a display feature. The platform also serves as a host environment for third party algorithms, with at least one external spatial biology vendor embedding its predictive biomarker technology into Concentriq. No named LIS integrations or standards were enumerated.

Operationally specific, with the host named and no location committed. The platform page states that Concentriq is built on Amazon Web Services with a 99.95 percent uptime service level agreement and regional redundancy, and gives scale: more than 6,000 cases an hour and more than 8 petabytes of research data. The August 2026 clearance of Concentriq AP-Dx (K261380) added cloud deployment. Implementation is described as a service covering deployment architecture, migration of images and data, scanner and systems integration and workflow redesign.

No region, residency commitment or tenant isolation model is published, and nothing states where images from a European or Canadian customer rest, although the clinical edition is sold under EU IVDR and a Health Canada license. Whether an on premises option remains is not stated.

One hosted service on AWS with location implied rather than committed.

No pricing is published, and the return side is. No license fee, per case or per slide rate, tier structure or worked example is published for any Concentriq edition, and commercial engagement runs through a demo request or contact form; a digital pathology buyer's guide sits behind a form. What is published is the upside: a 68 percent improvement in one day case turnaround at a named laboratory group, 60 percent faster workflow cycles at a contract research organization, a 40 percent reduction in time spent searching for data, and more than $1 million of new revenue at a named pathology group, plus a Best in KLAS 2026 award. A buyer sees a quantified return on an investment whose cost is undisclosed. How the platform is sold can be established in outline: a clinical edition for diagnostic laboratories and research editions for life sciences, with more than 230 AI applications on the platform whose pricing sits with their makers.

The sales posture is discoverable and no price is published.

Two customer bases, named, with the cleared scope stated precisely. The clinical edition serves high volume diagnostic laboratories for primary diagnosis on scanned surgical pathology slides from FFPE tissue, and excludes frozen sections, cytology and non-FFPE hematopathology specimens; the research editions serve life sciences organizations for biomarker research, clinical trials and companion diagnostics. The platform supports more than 45 scanners and formats including DICOM, SVS, TIFF, NDPI and iSyntax, and the August 2026 clearance added the Leica Aperio GT 450 DX. Dermatopathology is the evident depth of Proscia's own models, given the multi site classifier validation and dermatopathology workflow features, with biomarker scoring extending into oncology research.

Validation sits behind the cleared viewing scope and the earlier dermatopathology classifier; the current AI applications are research use only, and which scanners the cleared edition is validated with beyond those named in its clearances is not listed.