Proscia

A digital pathology platform with AI built into it rather than an AI product, and the company is deliberate about that framing. Concentriq is first an image management and workflow system for whole slide pathology, which is what high volume labs buy.

The AI layer is real and increasingly native, recommending stain panels and ancillary tests from what it sees in tissue and scoring image analysis biomarkers including PD-L1, HER2, and Ki-67 in the viewer, with the company describing an AI native architecture and domain specific vision, language, and multimodal models. The platform retains value without the AI; the AI does not exist without the platform.

Explicitly built around expert verification, and the company argues the position publicly rather than just implementing it. Outputs are described as backed by spatial, contextual, and quantitative evidence, transparent and interpretable and built for expert verification, with intelligence woven into high judgment workflows designed for experts in the loop.

In practice the pathologist reviews AI recommended stain panels and adjusts inline, and can modify regions of interest on biomarker scoring with results updating in real time. The company's CEO has publicly challenged the replacement narrative, arguing the opportunity is eliminating administrative burden rather than diagnostic expertise, which is consistent with how the product is scoped.

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.

This is the clearest answer to the central question of this axis located anywhere in the index, and it is made categorically rather than hedged. The company states that everything in the platform stays within the customer's organisation unless the customer gives express permission to de identify and share it to advance research, and that customer data is never used to train the models the platform runs.

That is the categorical form of the commitment rather than the narrower one excluding only identified data, and the distinction matters because de identified material derived from customer cases is still derived from customer cases, which is precisely the gap most vendors in this index leave open.

Publishing the exception alongside the promise is the second thing that earns the grade: the research sharing carve out is stated openly and framed as opt in by express permission rather than left for a buyer to find in a contract.

Supporting properties are consistent with the claim rather than merely adjacent to it, with data described as portable through open interfaces, customer built models executing as containerised workloads alongside the data with intellectual property retained by the customer, and controls covered by certified security and quality programmes.

Two residuals to obtain rather than assume: no retention period or deletion position at contract end is stated, and the platform carries a large catalogue of applications from other manufacturers, so confirm the no training commitment extends to hosted third party applications and not only to the vendor's own models.

Regulatory clearance for primary diagnosis is the anchor, supported by earlier multi site algorithm validation across more than 20,000 biopsy slides at named institutions including two university hospitals and two commercial dermatopathology laboratories. Deployment spans high volume diagnostic laboratories and large life sciences R&D organizations.

What is less visible publicly is quantified operational outcome data, meaning turnaround time, error reduction, or throughput figures from deployed customers, which the company references qualitatively rather than with numbers.

The clearest answer to the question this index puts to every AI vendor, and it is made categorically rather than hedged. Proscia states that everything in Concentriq stays within the customer's organisation unless the customer gives express permission to de identify and share it to advance research, and that customer data is never used to train the models Concentriq runs.

That is the categorical form of the claim, not the narrower one that merely excludes identified data, and the distinction matters because de identified material derived from customer cases is still derived from customer cases. Publishing the exception alongside the promise is the second thing that earns this grade: the research sharing carve out is stated openly and framed as opt in by express permission, rather than left for a buyer to discover in a contract.

Supporting properties are consistent with the claim. Customer data is described as remaining accessible and portable through open APIs, customer built models execute as containerised workloads alongside the data with intellectual property retained by the customer, and the underlying controls are covered by certified information security and quality programmes rather than asserted. What is not published is the rest of the lifecycle.

No retention period is stated, no deletion position at contract end is described, and no subprocessor list was located. The corporate privacy policy is published as a separate document and was not opened on this pass. A buyer should ask for retention and deletion terms in writing, and should confirm that the no training commitment extends to any hosted third party application running on the platform as well as to Proscia's own models, since the platform carries a large catalogue of applications from other manufacturers.

HIPAA is named explicitly and treated as a standard the platform is built against, which is more than most of this lane offers, but the instrument itself is never described. Proscia lists HIPAA alongside GDPR among the privacy standards Concentriq aligns to, and displays it beside its certified programmes.

The surrounding assurance is real and gives the alignment claim substance it would otherwise lack: information security and quality programmes certified under ISO 27001, ISO 13485 and SOC 2 Type II, a platform stated to be GxP ready for regulated research, and a change control practice in which every change is classified and validated by risk with a record of how each result was produced.

The customer base makes the obligation concrete rather than theoretical, including major United States health systems, a national reference laboratory network and a federal pathology centre. Two things hold it at B. Alignment is not certification, and this index consistently separates the two: a platform that aligns to HIPAA has not been assessed against the Security Rule in the way an audited programme has. And the business associate relationship is never addressed.

No business associate agreement is named, offered or described, and no statement sets out which obligations Proscia accepts as a business associate handling protected health information on a laboratory's behalf. A buyer should ask for the business associate agreement, and should establish whether it covers third party AI applications running on the platform or only Proscia's own services.

Among the strongest assurance positions in this lane, and the certifications are displayed rather than mentioned in passing. Proscia states that its security and privacy programmes are certified under ISO 27001, ISO 13485 and SOC 2 Type II, naming the report type rather than gesturing at certification, and presents them together with its HIPAA and GxP positions on the platform page where a buyer will actually find them.

Two operational commitments go beyond the certificates and are unusual in this category. A 99.95% uptime service level agreement with redundancy across regions is a contractual reliability position rather than an aspiration, which matters for a platform carrying primary diagnostic workload.

And the company states that every change is classified and validated by risk with a clear record of how each result was produced, which is result level provenance and is directly relevant to a laboratory that may later need to reconstruct how a given case was analysed. The scale being protected is substantial, with customers managing more than 8 petabytes of research data and 6,000 cases an hour. Two limits are worth stating rather than treating as disqualifying.

No certificates are published as documents with their scope and expiry, so the claims cannot be verified without contacting the company, and no trust centre, penetration testing statement or subprocessor list was located. GxP readiness applies to regulated research environments and should not be read as an information security attestation, though it is meaningful to pharmaceutical customers for separate reasons. Ask for the current SOC 2 report and its trust services criteria, the ISO 27001 certificate and scope, and the subprocessor list.

Cleared for primary diagnosis, which is the highest bar in digital pathology, and the company draws the boundary with unusual discipline. Concentriq AP-Dx is FDA 510(k) cleared in the United States and CE-IVDR marked in the EU for primary diagnosis. Critically, the company states plainly that the separate research edition is a Research Use Only product not for use in clinical diagnosis, and directs clinical buyers to the cleared edition instead.

Publishing that distinction prominently rather than blurring it is exactly the practice the index's pathology category calls for, since Research Use Only is a permitted use restriction that vendors frequently obscure.

No formal governance program was located, but the transparency posture is substantive: outputs are stated to be interpretable and evidence backed with granular overlay tools letting pathologists confirm results visually, and the early validation deliberately spanned four independent laboratories rather than a single site. No published analysis addresses performance variation across scanner platforms, staining protocols, or patient demographics, which are the known drift sources in computational pathology.

The strongest validation on this record describes a component rather than the product being sold, which is a distinction this index has now recorded several times and which a buyer should apply here. The dermatopathology classifier was validated in a multi site study across named academic and commercial laboratories using more than twenty thousand patient biopsy slides, and that is a serious piece of work: multi site matters because a pathology model trained in one laboratory meets different scanners, stains and preparation practices in another, and named sites let a reader judge whether their own setting resembles them.

What is absent is anything equivalent for what ships now. Current model performance figures and validation methodology for the shipping features are not published at that level of detail, so a buyer purchasing today is relying on the reputation of an earlier validation rather than on evidence about the software in front of them.

The architecture is described specifically as domain specific vision, language and multimodal models reasoning across images and data with clinical context unified in one layer, which tells a reader the shape without telling them the performance. No warranty, indemnity or remediation commitment was located. Ask which shipping features have published validation and which do not, and for performance on your scanner and stain protocol.

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.

The most operationally specific deployment position in this lane, published in numbers rather than adjectives. Scale is stated and it is large: customers manage 6,000 cases an hour and more than 8 petabytes of research data, the platform is used by over 4,000 hospitals and clinics and by 16 of the top 20 pharmaceutical companies, and named institutions include major health systems, national reference laboratories and a federal pathology centre.

Reliability carries a published commitment rather than a claim, with a 99.95% uptime service level agreement and redundancy across regions. Interoperability is the strongest in the category: support for more than 45 scanners and every major image format including DICOM, native bi directional laboratory information system and laboratory information management system integrations, open APIs, and a managed compute service that will run a customer's own containerised models alongside their data with intellectual property retained.

Implementation is described as a service, with a professional services team stated to have taken hundreds of laboratories through digitisation, covering deployment architecture, migration of existing images and data, systems and scanner integration and workflow redesign. One regulatory detail belongs on this axis because it governs what can actually be deployed for clinical use.

Concentriq AP-Dx is FDA cleared for primary diagnosis in the United States only in combination with a specific named scanner, is licensed in Canada and CE marked under the EU IVDR, and is expressly not intended for frozen sections, cytology or non FFPE haematopathology. Concentriq AP and Concentriq LS are research use only.

The company also states that buyers must contact the individual AI application manufacturers for regulatory authorisation status, so an application's clearance does not travel with it onto the platform. Held at A on deployment substance while noting the one gap: regional redundancy is mentioned but no data residency commitment naming supported regions was located.

No pricing is published. No licence fee, no per case or per slide rate, no tier structure and no worked example was located for any Concentriq product, and commercial engagement runs through a demo request or a contact form. What is published instead is the return side, in specific and quantified form: a 68 percent improvement in one day case turnaround at a named laboratory group, 60 percent faster workflow cycles at a contract research organisation, a 40 percent reduction in time spent searching for data, and more than one million dollars of new revenue attributed to a named pathology group.

That combination is the upside published and price gated asymmetry this index tracks across several categories, and it is well developed here. A buyer is shown a quantified return on an investment whose cost is not disclosed, which makes the ratio impossible to complete without entering a sales process. Two partial credits. A digital pathology buyer's guide is offered, though behind a form.

And independent customer research is cited, including a regional Best in KLAS award and a reported finding that all customers interviewed would buy again, which is validation of satisfaction rather than of value for money. A buyer should ask for the charging basis and specifically whether it is per case, per seat, per scanner or by storage volume, since a platform whose customers manage petabyte scale data may price on a dimension that grows independently of clinical benefit. Ask also how third party AI applications running on the platform are charged, and by whom.

Two distinct customer bases served by two editions: high volume clinical diagnostic laboratories using the cleared product for primary diagnosis, and life sciences R&D organizations using the research edition for biomarker discovery and clinical trials. Dermatopathology is the evident depth given the validated classifier and derm specific workflow features such as automatic slide orientation, with biomarker scoring extending into oncology. This is a laboratory product; it has no provider or hospital clinical workflow surface.