Rayscape

The models are the product. Rayscape CXR detects and localizes 148 findings across 17 clinical classes on chest X-rays, including pneumothorax, pleural effusion, consolidation, lung lesions, tuberculosis, cardiomegaly and atelectasis, with bounding boxes and probability scores, generates bone suppressed and bone subtracted views, calculates the cardiothoracic ratio, and flags studies as normal or abnormal for prioritization, according to the company's AWS Marketplace listing. The company also sells lung nodule detection on CT. There is no hardware, archive or services layer underneath; remove the models and nothing is left to sell.

Assistive detection with the radiologist deciding. Rayscape CXR marks findings with bounding boxes and probability scores, flags each study as normal or abnormal, and prioritizes abnormal studies, and its reporting engine converts detections into draft report text for the radiologist to review and validate. The company has also coauthored research on how textual and visual explanations of an AI result change reader decisions, which bears directly on how much a reader defers to the output.

The structure is described and one part is missing: the thresholds behind the normal and abnormal flag are configurable by site and not published, and nothing states how a study flagged normal is handled in a busy worklist or what happens after a missed finding. For a product that flags normal studies, that routing is the part a buyer needs in writing.

The approach is described, as deep learning detection that localizes findings with probability scores, and what the model produces is listed in detail: 148 findings across 17 clinical classes, bone suppressed and bone subtracted views, and an automatic cardiothoracic ratio.

The company has also published research, including a survey of the capabilities and evidence coverage of CE marked lung nodule products and a study of how explanation format affects clinical decisions, and it states openly that tuberculosis detection depends on thresholds tailored to the setting. What is not published is the model architecture, the training data (size, sources, countries, equipment makers), per finding accuracy across the 148 findings, or versioning and update practice.

The chain is short by the company's own account. The models are Rayscape's own, and the company has stated that patient data is never transmitted outside the customer's institution and is not shared with any external party, which, where it holds, is a deployment in which the transfer to the vendor does not occur. Two things keep it from being settled.

The company's AWS Marketplace listing describes the product as software as a service, deployed on premises or in the cloud, and a service run by the vendor is hard to square with data never leaving the institution; which applies to which deployment is not stated. And for any cloud path, no hosting provider, region or subprocessor is named; the listing states the product is not deployed on AWS.

The strongest evidence point is independent: tuberculosis detection was assessed in a study of commercially available computer aided detection software within a South African tuberculosis prevalence survey, published in The Lancet Digital Health in 2024, a population screening setting rather than a curated retrospective cohort. The paper carries the product's figures and how they compared with other software at the thresholds the survey used.

Beyond tuberculosis, the product is sold on 148 findings, and per finding performance is not published; the company reports clinical use at more than 300 sites in 18 countries, which is scale of use rather than evidence of benefit, and figures such as reduced reading time come from customer stories without method. A conference poster on pneumothorax detection (ECR 2023) is the one other published evaluation.

The company has made a categorical commitment: patient data is never transmitted outside the customer's institution and is not shared with any external party, with adherence to European data protection law stated alongside it. Where that holds it removes the vendor from holding patient data, and with it the question of whether customer images train the models. No retention schedule sits behind it, and three points remain open.

The company's AWS Marketplace listing describes the product as software as a service, which sits uneasily with data never leaving the institution, and nothing says which deployments the commitment covers. Support, diagnostic and incident access usually carries exceptions. And where a site uses a cloud hosted archive, the institution's boundary is no longer physical. Safety engineering, meaning guardrails and how a safety event is handled, is not published.

Rayscape is a Romanian company with CE marking and no FDA clearance, so its data protection position is European, and it has stated one: adherence to European data protection law and a commitment that patient data never leaves the customer's institution.

The service privacy document or data processing agreement behind that position would sit on the company's own site, which builds its pages in the browser; none is named in the company's AWS Marketplace listing, which describes the product as software as a service and makes no data handling statement.

A software as a service product and a commitment that patient data never leaves the institution do not obviously fit together, and the documents are where a buyer would see which deployments the commitment covers. If a US hospital were to use the product, business associate terms would attach.

The company states that it operates under ISO 13485 quality management and is certified under the EU Medical Device Regulation, with conformity audits by an outside body. Those certify the quality system and the device, not how the company operates information security, and a notified body's review of a device's technical documentation, including its cybersecurity risk analysis, reviews the manufacturer's own submission.

Neither its AWS Marketplace listing nor its other published material names a SOC 2 report, ISO 27001 certificate, penetration test or trust center. The privacy page, the footer and any security page on its own site may settle the question either way. The cybersecurity section of the technical documentation can be requested.

The regulatory position is European. The company states that its products are CE marked under the EU Medical Device Regulation, with a quality system under ISO 13485, and its AWS Marketplace listing describes Rayscape CXR as CE marked. No FDA clearance is listed for the company or its products, so none of them is cleared for clinical use in the United States; the marketplace listing, open to buyers anywhere, does not claim one.

The site gives no device class, notified body, certificate number or certified intended use. The largest tuberculosis screening programs are in high burden countries, where CE marking and World Health Organization aligned evidence often matter more than FDA clearance.

An independent tuberculosis study in a South African prevalence survey (The Lancet Digital Health, 2024) tests the product in a population and disease burden very different from typical training data. The full paper may include results by group, and it is unclear whether the company cites it. The company states openly that tuberculosis thresholds need tailoring to the setting, which acknowledges that performance varies by context without publishing by how much.

It publishes no results by sex, age, HIV status, prior tuberculosis, equipment maker or disease prevalence, and names no governance framework. Chest X-ray classifiers trained on public datasets have underdiagnosed female, Black, Hispanic, younger and Medicaid insured patients (Seyyed-Kalantari and colleagues, Nature Medicine, 2021).

Rayscape CXR shows each finding with a bounding box and a probability score, and its draft report text is reviewed and validated by the radiologist before sign off, so a reader can see what the model marked and why. No warranty, indemnity or remediation commitment is named, and the company's material on record gives no error rate with its method and denominator for its 148 findings.

The independent tuberculosis study in The Lancet Digital Health (2024) reports how the product performed in a South African prevalence survey; its full text sits behind a browser check on the publisher's host, and the company's own site, which builds its pages in the browser, is where a citation of it would appear. If the company cites it and it gives sensitivity and specificity with denominators, that puts an error rate on the record.

Rayscape CXR integrates with the archive over DICOM, on premises or with a cloud archive, returning annotated images and structured outputs inside the native radiology viewer without a separate workspace, and sites can configure outputs, thresholds and visualization to their own reporting standards.

It is listed in the Blackford Analysis imaging AI portfolio, a named orchestration platform, and the company reports clinical use at more than 300 sites in 18 countries, with customer stories on its site describing emergency and outpatient deployments. That is standards support with a named platform behind it. No integration with a named record system, RIS vendor or API is documented, and no HL7 or FHIR path for results is described.

Two options are stated. The company's AWS Marketplace listing describes Rayscape CXR as software as a service deployed on premises or in the cloud, and states that it is not deployed on AWS, and the company has stated that patient data never leaves the customer's institution, which describes the on premises path.

For the cloud or service path the listing names no hosting provider, region, residency commitment or tenant isolation, and the company's own site, which builds its pages in the browser, is where those would be stated. An on premises option matters for institutions and screening programs that cannot export imaging, and for them residency is their own.

Clinical use carries no published price that could size a purchase. The company's site has a pricing and plans page, and its AWS Marketplace listing for Rayscape CXR shows a free subscription with no end date alongside a 30 day money back guarantee on the current annual term, which suggests a paid annual contract behind the listing without stating its price or unit. The product can be procured through the marketplace and directly. The company is early stage, having reported a seed round in March 2024.

Thoracic coverage across two modalities, with validation behind part of it. On chest X-ray the product claims 148 findings across 17 classes for PA and AP studies, plus tuberculosis detection; on CT it detects lung nodules with comparison against a patient's prior scans to track change over time.

The tuberculosis use has independent validation in a South African prevalence survey (The Lancet Digital Health, 2024), a population screening setting; the broader findings set and the CT product have no published per finding validation. The company reports clinical use at more than 300 sites in 18 countries, and its customer stories describe emergency and outpatient deployments. Because no FDA clearance is listed, the products are not offered for clinical use in the United States.