PictorLabs

The models are the product. Every PictorLabs product returns an image that a deep learning model generates: a virtual H&E from an unstained section, several virtual stains from one autofluorescence scan, virtual special stains from an existing H&E slide, or a stain converted between scanners and imaging methods. PictorLabs sells no scanner, stain or laboratory service. Without the models there is nothing to deliver.

Ask which model version produced each virtual stain in a study, and whether models are retrained for a customer's tissue or scanner.

Checking outputs is left to the user, with no control described. PictorLabs's terms say it makes no representations about the accuracy of AI generated outputs, and that users rely on them at their own risk and must verify them. Every product carries a research use only label.

The published studies used blinded pathologist review. Nothing describes how a customer should check a virtual stain against a chemical one, when a stain should not be relied on, or how quality is controlled before delivery.

Ask what quality checks run on each virtual stain before delivery, and what reference staining PictorLabs recommends for validating a study.

The approach is published in method papers, without product versions or update practice. PictorLabs's products rest on deep learning methods its founders published. They turn label free autofluorescence or brightfield images, or chemically stained sections, into other stains, and each product names its inputs and outputs.

The models behind each product are not named or versioned, and their training data is not described on PictorLabs's pages. No release notes or change notices are published.

Ask for the model version and training data behind each stain, and how customers are told when a model changes mid study.

PictorLabs builds its own staining models from methods its founders published, and no outside model provider is named or implied. The on premises option keeps images in the customer's environment.

For the cloud option, no hosting provider, region or subprocessor list is published. The privacy policy says personal information is shared with service providers without naming them.

Ask for the cloud host and the subprocessors that handle images in the cloud service.

Peer reviewed research led by the founders supports the method, without independent validation of the products sold. A 2019 paper in Nature Biomedical Engineering turned autofluorescence scans of unstained tissue into H&E, Jones and Masson's trichrome stains across several organs, with blinded review by board certified pathologists. A 2021 paper in Nature Communications turned H&E kidney biopsies into virtual special stains for kidney disease other than cancer.

Later papers cover label free HER2 staining of breast tissue (2022) and virtual H&E in lymphoma (Journal of Hematopathology and International Journal of Surgical Pathology, 2024). Conference work includes a 2024 comparison of virtual and chemical H&E for lymphoma diagnosis with Johns Hopkins, UCLA and the University of Maryland, and a neurotoxicity study with Charles River.

PictorLabs's evidence page lists these studies without figures. The papers do not name product versions, and none is an independent evaluation of ClearStain, DeepStain, ReStain or RetroStain as sold.

Ask for agreement figures between virtual and chemical stains on your tissues and scanners, and which product version each study used.

PictorLabs gives general privacy assurances, without answers on retention or training use. It displays SOC and CCPA compliance marks and offers an on premises option for full data control. It publishes nothing on how long images sent to its cloud are kept, whether customer images train or refine its staining models, or how training tissue was sourced and de identified.

Its privacy policy covers account and website data, and allows de identified or aggregated data to be shared with third parties. No safety engineering for the models is published, such as how a virtual stain that invents or misses structure is caught.

Ask how long cloud processed images are kept, whether customer images train any model, and how staining artifacts are caught before results are delivered.

PictorLabs's privacy policy reaches people who hold accounts with the service, and takes no position on patient data. The policy, effective June 2026, covers website visitors, account holders, event participants and business partners. It says aggregated or de identified data may be shared with third parties.

It does not address tissue images or any patient information in them, and no business associate status, scope position or agreement is published. The products are sold for research, where images are often de identified, but nothing published says so.

Ask whether images sent to the cloud service may contain protected health information, whether PictorLabs signs business associate or data processing agreements, and how the on premises option changes that.

PictorLabs displays a SOC compliance mark, without the report type, scope or date. Its site carries an AICPA SOC mark and a CCPA compliance mark in the footer. The AICPA mark indicates an audit of controls by an independent accountant.

PictorLabs does not say whether it is a SOC 2 report, whether it is Type I or Type II, what it covers or when it was issued. No trust center or penetration testing statement is published.

Ask for the SOC report itself, its type, period and scope, and whether it covers the cloud staining service.

PictorLabs makes no device claim, and its products are labeled for research. Each product carries a research use only, not for use in diagnostic procedures label. Its evidence page says its virtual staining is not cleared or approved for clinical diagnostic use, and its terms say the technology is not approved by the FDA for clinical use. The FDA's databases list no PictorLabs device.

Its published research includes diagnostic comparisons in lymphoma and HER2 breast tissue. These describe where the method could go, not what is sold today.

Ask whether any product is planned for a clinical pathway, and how research results made with virtual stains should be handled in regulated studies.

PictorLabs lists studies across tissues and stains, but publishes no results by group and no governance process. The 2019 paper is described as validated across organs and stain types, and RetroStain is offered to even out color across scanners.

Even so, PictorLabs's pages set out no results compared across tissue types, scanners, laboratories or staining protocols. No governance framework, model evaluation process or monitoring after release is described.

Ask for agreement figures by tissue type, scanner and laboratory, and how a new stain on the biomarker menu is evaluated before release.

Responsibility for outputs sits with the user. PictorLabs's terms say it makes no representations about the accuracy of AI generated outputs, that users rely on them at their own risk and must verify them, and that disputes go to arbitration.

A virtual stain can be checked against a chemical stain of the same tissue, which is how the published studies were run. But no error rate for a product as sold, warranty, reprocessing commitment or indemnity is published.

Ask what PictorLabs does when a delivered virtual stain is found to misrepresent tissue, and what a commercial agreement adds to the website terms.

No clinical record integration exists, and no research system integration is described. PictorLabs's products work on whole slide images in research pipelines, on premises or in the cloud.

Its home page shows Grundium, Hamamatsu, Proscia and Pramana among its partners, without saying what each integration does or whether it is in production. No image management system, laboratory system, file format or interface is documented.

Ask which scanners and image management platforms the products read from and write to in production, and in which file formats.

Two deployment options are named, with no location committed. PictorLabs offers on premises deployment for full data control or existing imaging pipelines, and a cloud option for scaling and for studies across several sites.

Nothing published says which cloud provider or region hosts the cloud service, where images rest, or how customers are kept apart.

Ask for the cloud provider and region, where images and outputs are stored and for how long, and how customer data is isolated.

No price is published, but the outline of how the products are sold can be worked out. PictorLabs describes four software products for research use, run on premises or in the cloud. A biomarker menu is available on request, and access starts with a demo request.

Nothing published says whether customers pay per slide, per stain, per study or by license, or what an on premises deployment costs.

Ask for the unit of charge for each product, the license terms for on premises use, and whether new stains on the biomarker menu are priced separately.

PictorLabs names its stains and settings, with peer reviewed validation behind part of them. The products cover virtual H&E, special stains (trichrome, reticulin, Jones, PAS) and stains that resemble immunohistochemistry markers (PanCK, CD45), with more on a biomarker menu on request. Inputs include brightfield, autofluorescence, H&E, immunohistochemistry and multiplex immunofluorescence images. Published work covers several organs, kidney biopsies, breast tissue for HER2 and lymphoma.

PictorLabs names drug development, preclinical safety and toxicology studies among its uses, with a neurotoxicity study run with Charles River, which is why it is also listed under drug discovery. All use is research only, and validation for markers on the request menu is not published.

Ask for validation of each stain on the tissue types in your study before relying on it.