Artrya

AI does the heavy lifting inside a physician-driven reading tool. Salix is cleared as a web-based interactive tool for cardiologists and radiologists to view and analyze coronary CT; its 510(k) summary for Salix Coronary Plaque (K251837) describes machine learning for centerline extraction, vessel labeling, lumen and vessel wall segmentation and calcium scoring, with plaque volumes computed from Hounsfield unit ranges between the lumen and vessel wall, and physicians editing the output to prepare the report.

Customers pay a monthly subscription for the Coronary Anatomy reading platform and a fee per scan for plaque analysis. The automation is what makes the tool fast, but a coronary CT reading workstation still has value without it. AI central to a broader offering.

The physician reviews and can edit every output; where the tool should not be used is not published. Salix returns an automated analysis in minutes without an external reading team, and the physician starts from that output and edits segmentation, stenosis and plaque as needed before reporting; the 510(k) summary for Salix Coronary Plaque (K251837) describes full manual editing and tested it with routine clinicians rather than core lab analysts.

The Indications for Use say Salix is an adjunctive tool that does not replace a professional's review or judgment, is for trained users, and warn that interpolated views can make tissue look healthy where small pathology exists. The company markets Salix for rapid discharge of chest pain patients in emergency and primary care; the cleared indication is adjunctive analysis of CCTA, and no published limits say which patients or scans fall outside it. The review and edit path published, the limits and thresholds missing.

The machine learning functions are named and validation data is kept apart from training; update practice and training detail are not published. The Salix Coronary Plaque 510(k) summary (K251837, version 1.0.0) identifies which outputs use machine learning (centerline extraction, vessel labeling, lumen and vessel wall segmentation, calcium scoring) and says plaque volume is computed from Hounsfield unit ranges between the lumen and vessel wall, reporting low attenuation plaque, spotty calcification, positive remodeling and the napkin-ring sign.

It states that validation data came from US centers that contributed nothing to the training data. The company says the algorithms were built from thousands of scans from institutions in the US, Australia and Canada. Nothing is published on the training data's size and composition, on how model updates are validated and released, or on any derivation of the proprietary Plaque Dispersion Score the company is studying.

A short chain with the host named. Artrya builds its own models, cleared in its own name, from scans contributed by institutions in the US, Australia and Canada, so no outside model provider sits between a CT study and a result, and the 510(k) summary for Salix Coronary Plaque (K251837) names Amazon Web Services as the hosting provider. The AWS regions, a subprocessor list and the terms on which contributing institutions supplied training scans are not published. The privacy policy separately allows disclosure to a third party that may combine data for anonymized consumer insights, with no clinical carve-out.

Measurement accuracy is validated for clearance; the clinical claims the company leads with are not. The Salix Coronary Plaque 510(k) summary (K251837) reports a retrospective multi-reader study of 103 patients from seven US centers in which routine clinicians using Salix agreed closely with expert readers (stenosis within one CAD-RADS category in 95.8 percent of vessels; total plaque ICC 0.96), which shows the measurements are sound.

The home page leads with outcome claims: rapid discharge of chest pain patients with confidence there are no plaque features that could cause future events, fewer unnecessary admissions and lower costs. No published study tests those. SAPPHIRE, a retrospective study across six US health systems of the company's Plaque Dispersion Score with a focus on women, has not reported. Evidence that does not reach the headline claim.

The posture is far from empty, but it is ambiguous. Three product specific commitments are stated, and they are firm ones: users are contractually told not to upload personal information about any person into the software except as agreed in the customer agreement; anything that is uploaded will be used only to administer that user's use, which on its face excludes research and model development; and an account holder can request erasure of all data associated with their account through the Privacy Officer.

Records no longer required are deleted or permanently de identified. What is missing is scope and duration. No retention period is given, and because the default instruction is not to put personal information into the product at all, it is unclear which document actually governs an identified coronary CT study.

The same policy separately permits disclosure to a third party who may combine data to develop anonymized consumer insights, and lists advertising and marketing among its purposes, with no clinical carve out. Nor is it stated whether studies are deidentified before upload or after.

The governing regime named is the Australian Privacy Act 1988 and the Australian Privacy Principles, with a Privacy Officer as the contact point. HIPAA and the business associate agreement are absent entirely. The gap is pointed rather than theoretical: the US product is FDA cleared under K243038 and the company publishes its expected US reimbursement pathway, so American clinical deployment is central to the commercial plan while the only published privacy instrument is bound to Australian law. The Australian Privacy Principles are a real regime, and this is not an absence of governance, but they are not the regime a US health system is surveyed against.

Controls are described with nothing independent behind them. Artrya's privacy policy describes physical, administrative, personnel and technical measures in general terms and says servers holding personal information are encrypted and records no longer needed are deleted or de-identified; it also says information transmitted to the company is sent at the user's own risk, without separating website contact data from clinical imaging. Salix runs on Amazon Web Services.

No SOC 2 report, ISO 27001 certification, HITRUST, independent penetration test or trust portal is published, and the customer announcements mention only customary security and privacy terms. Controls asserted with no outside check.

Each module's status is identifiable, and the company says which are not yet cleared. The Salix reading platform (Salix Central, K243038, named in the site footer) was cleared in March 2025, and Salix Coronary Plaque version 1.0.0 was cleared on 20 August 2025 (K251837; 21 CFR 892.2050, Class II) for quantifying and characterizing coronary plaque and stenosis and calcium scoring on cardiac CT, as an adjunctive tool that does not replace a professional's diagnostic decision-making.

Salix Coronary Anatomy is listed on the Australian Register of Therapeutic Goods (ID 347719); availability in the EU and UK is by inquiry. Salix Coronary Flow, an FFR-CT module, is described as in validation ahead of a 510(k) submission and is not available.

The validation cohort's make-up is published, and the company's study is designed around women; no results by group exist yet. The Salix Coronary Plaque 510(k) summary (K251837) describes its 103 validation patients by sex (58 women, 45 men), age (23 to 84), self-reported race (57 percent White, 22 percent Black, 12 percent Asian) and ethnicity (13 percent Hispanic or Latino), across four scanner makers at seven US centers, but reports performance only for the whole group.

SAPPHIRE, Artrya's study across six US health systems, is designed with a stated focus on coronary disease in women, on the rationale that most women who die of it had no prior symptoms; it has not reported. No AI governance framework or bias policy is published. Specific disclosure of who the product was tested on and a study built around the most at risk group, without group results.

An error rate for the machine learning functions, with its method and denominators, is in the company's own filing; nothing stands behind the output. Artrya's 510(k) summary for Salix Coronary Plaque (K251837, cleared 20 August 2025) reports standalone testing of the machine learning outputs against reference segmentations by SCCT Level III cardiologists and radiologists, on data from seven US centers that contributed nothing to training: calcium scoring on 302 non-contrast series (Pearson 0.958), and centerline extraction, vessel labeling and wall segmentation on 107 contrast studies (lumen Dice 0.90, vessel wall Dice 0.90, labeling F1 0.83).

A separate multi-reader study of 103 patients found clinicians using Salix agreed with expert readers on stenosis within one CAD-RADS category in 95.8 percent of vessels, with plaque volume ICC from 0.96 (total) to 0.61 (low attenuation). Physicians can edit every output. No warranty, indemnity or remediation commitment is published, and the proprietary Plaque Dispersion Score has no published derivation.

Integration into a named customer's PACS and medical record is reported, with the systems themselves unnamed. Artrya's announcement of its Tanner Health contract says Salix Coronary Anatomy was integrated into Tanner's PACS and electronic medical record and is in clinical use across its five hospitals; Northeast Georgia Health System and Cone Health are further customers.

Salix reads DICOM from any CT scanner, and new modules such as Coronary Plaque switch on inside the same interface without a separate integration project. No PACS or record vendor, interface standard or marketplace listing is published. Integration with named deployments, systems unnamed.

A single cloud service on a named host, with no location committed. Salix is web-based software delivered as a service on Amazon Web Services (K251837), returning results within minutes inside one interface, with no on-premise option. Artrya operates in Australia and the US, and no AWS region, data residency commitment or tenancy model is published for either. One hosted option, location implied.

The pricing structure and one contract's value are published in the company's stock exchange filings; the per scan fee is not. Artrya's ASX announcement of its first US customer states a five year SaaS license for Salix Coronary Anatomy at a monthly subscription fee, valued at US$0.6 million over 60 months for Tanner Health's five hospitals and 30 practices, with a fee per scan added for Coronary Plaque and, once cleared, Coronary Flow.

Further US customers are reported on the same subscription plus per scan structure. The Category I CPT payment for plaque analysis (US$950 per assessment from January 2026) is what a provider can bill, not Artrya's fee. A pricing basis with a figure that lets a buyer size the subscription before calling, short of a price list.

Coverage is named for coronary CT with measurement validation behind it; the emergency and primary care rule-out use the company markets goes further than the evidence. The cleared indication covers adjunctive analysis of CCTA for patients evaluated for known or suspected coronary disease and calcium scoring on non-contrast cardiac CT, on any CT scanner. The validation in K251837 spans seven US centers, four scanner makers, adults aged 23 to 84 and routine clinician readers.

The home page leads with rapid assessment and discharge of chest pain patients in emergency and primary care, a setting where no validation is published. Deployments run across Tanner Health's five hospitals and further US health systems. The narrow disease focus is not the limit. Named coverage with validation behind part of it.