ArteraAI Prostate

Performance Metrics

Clinical Evidence

ArteraAI Prostate has an unusually strong evidence pedigree for an AI device, and the reason is methodological rather than promotional: its models were developed and tested on archived diagnostic tissue from completed phase 3 randomised controlled trials, where treatment was allocated at random and patients were followed for a decade or more. Randomisation is what separates a prognostic marker (this patient will do badly) from a predictive one (this patient will benefit from this therapy), and it is the predictive claim that makes this device distinctive. None of the studies is a randomised trial of the device: each is a post-hoc biomarker analysis of trial tissue, which is why the evidence grade here is prospective rather than RCT. The foundational architecture was reported by Esteva and colleagues in npj Digital Medicine in 2022, training a multimodal deep-learning model on digital histopathology and clinical data from men enrolled in NRG/RTOG phase 3 trials. The predictive result followed in NEJM Evidence in 2023 (Spratt et al.): the model was developed on 5,727 patients from five phase 3 randomised trials of radiotherapy with or without androgen-deprivation therapy, locked, and then validated on the NRG/RTOG 9408 trial (n=1,594). In that validation cohort 543 men (34%) were model-positive, and among them ADT significantly reduced distant metastasis; among the 1,051 model-negative men (66%) ADT showed no benefit. That is the first AI-derived biomarker of ADT benefit to be validated on randomised data. A prognostic companion analysis (Tward et al., JCO Precision Oncology 2024) pooled 9,787 patients across eight NRG Oncology trials and found the MMAI risk grouping stratified 10-year metastasis risk more sharply than NCCN risk groups, particularly within the heterogeneous high-risk band. Two features of the evidence temper the enthusiasm and both deserve emphasis. First, the great majority of these analyses are authored or funded by the manufacturer and are post-hoc, ancillary studies of archived trial tissue rather than biomarker-primary prospective trials — a legitimate and powerful design, but not the same as a device tested prospectively against its intended clinical use. The most independent validation to date is external: a UCL Cancer Institute and MRC Clinical Trials Unit-led study in Lancet Digital Health (2025; Parker et al., with Artera co-authors) applied the locked model to 3,167 patients from four phase 3 STAMPEDE trials in advanced disease and found it retained prognostic value there, strengthened when combined with disease burden. Second, a prospective, real-world validation does exist (Bjartell et al., Clinical Cancer Research 2025; n=143, Sweden) but in a radical-prostatectomy cohort — a different treatment context from the radiotherapy trials the predictive model was built on, so its predictive readout should not be read across to surgical patients. The NCCN recommendation is Category 2A — uniform consensus, not the highest evidence tier — and applies to localised disease.

Inside the Auris+ Listing

Regulatory Approvals

Safety Record

No recall, FDA safety communication, or adverse-event signal for ArteraAI Prostate was found in indexed public sources as of August 2026 — read as "none found in public reporting" rather than an exhaustive audit, since the FDA MAUDE and enforcement databases were not directly queryable from this environment. The device's harm surface is bounded by what it is: a risk score delivered alongside, not instead of, standard pathology and clinical assessment. It does not read the slide in place of a pathologist, does not make a cancer diagnosis, and does not remove any step from the diagnostic pathway. The residual risk is the one common to all treatment-selection tools — that a score is over-weighted against clinical judgement — which is precisely why the FDA authorisation and the NCCN guidance both frame it as one input among several, used with NCCN risk group, individual patient factors, and preference.

Intended Use & Indications

ArteraAI Prostate is a software-only medical device intended to analyse scanned histopathology whole-slide images from treatment-naïve prostate core-needle biopsies prepared from formalin-fixed paraffin-embedded tissue and stained with haematoxylin and eosin. Combining the image features with routine clinical inputs — patient age, PSA, and tumour stage — the multimodal model returns a continuous risk score that prognosticates long-term oncological outcomes in men with non-metastatic prostate cancer. The FDA-authorised role is prognostic risk stratification: the De Novo indication covers 10-year risks of distant metastasis and prostate-cancer-specific mortality in men aged 55 or older without clinically or pathologically defined metastases who are candidates for curative-intent management (surgery, radiotherapy with or without systemic therapy, or active surveillance), used alongside established clinicopathological factors, National Comprehensive Cancer Network risk group, and patient preference to inform shared decision-making about the intensity of treatment. In the localised-disease setting a distinct predictive readout, validated in randomised-trial data, estimates whether an individual is likely to derive benefit from adding androgen-deprivation therapy to radiotherapy — the first AI-derived predictive biomarker of that treatment benefit to be validated on randomised cohorts. The device does not render a cancer diagnosis, does not read the slide in place of a pathologist, and does not recommend a specific therapy; it stratifies risk to support the treating clinician's judgement.