Koios DS

Performance Metrics

Clinical Evidence

Koios DS has an unusually layered evidence base for a catalogue lean-to-hero graduation — manufacturer pivotal studies underneath, independent external validations on top, and it is the rare AI ultrasound tool evaluated across two organs. The breast pivotal is Mango and colleagues (Radiology venue AJR, 2020): a multicentre retrospective multireader study in which 15 physicians read 900 breast lesions (roughly half benign, half malignant) with and without the software. Mean reader area-under-curve rose from 0.83 reading ultrasound alone to 0.87 with decision support. It is a competent reader study, and it is also the shape pivotal studies usually take: enriched, retrospective, and closely tied to the manufacturer (an earlier data-presentation study, Barinov 2019, was authored from inside Koios). The thyroid clearance rests on a second multireader study — 15 readers, 650 nodules, a mean reader AUC gain of 0.083 — that was reviewed by the FDA but, as far as could be established this run, never journal-published; it lives in the 510(k) summary rather than the literature. What lifts Koios above a one-study entry is the independent work. A prospective single-centre European study (European Journal of Radiology, 2025; 222 patients, 226 lesions) found the software significantly improved the AUC of every reader, with the largest gain for the junior radiologist (0.786 to 0.864) and a reduction in unnecessary biopsy recommendations. Two independent thyroid cohorts agree on direction: a Valladolid group (Thyroid, 2024; 172 patients) found the system reclassified most mildly-suspicious TI-RADS 3 nodules to lower risk, and a Duke group (Thyroid, 2024; 649 nodules) found performance best when the radiologist read was combined with the software's adapter score, cutting unnecessary fine-needle aspirations. The honest counterweight is specificity. An independent Turkish comparison (European Journal of Breast Health, 2025) reported the software used standalone at 92.5 percent sensitivity but only 35.3 percent specificity — below an experienced radiologist reading unaided — and low specificity recurs as the caveat across the thyroid literature too. The consistent finding is that Koios DS is a decision aid that raises the floor for less experienced readers, not a standalone diagnostician; the published record supports the former claim and actively cautions against the latter.

Clinical Pulse

Verified clinician sentiment

Verified clinician reviews launching soon · Apply to contribute →

Reviews are signed by clinicians verified against their regulator's public register.

Inside the Auris+ Listing

Regulatory Approvals

Safety Record

The one recall on record is a resolved documentation matter, not a clinical-harm event. In December 2019 the pre-rebrand product line — Koios DS Breast series 1.0, the ClearView-era software behind the 2016 clearance — was subject to a Class 2 recall (Z-0820-2020) after an internal audit found affected versions had been marketed without sufficient documentation of certain required quality procedures. Urgent Medical Device Recall letters went out on 31 December 2019 and the recall was terminated on 7 May 2020. It concerned quality-management documentation, not device performance or a patient-safety signal, and it predates every current-generation clearance. No FDA safety communication or warning letter for Koios Medical was found in indexed public sources as of July 2026. The MAUDE adverse-event database could not be queried directly from this environment, so the absence of adverse-event reports is "none found in public reporting" rather than an exhaustive audit. As an adjunct that operates on a physician-selected lesion and removes nothing from the worklist, the device's harm surface is bounded by the cleared instruction that management decisions must not rest on its output alone.

Intended Use & Indications

Koios DS is prescription image-analysis software used by trained interpreting physicians as an adjunct to diagnostic ultrasound. For breast, it assists in analysing ultrasound images of adult patients with soft-tissue breast lesions referred for further diagnostic examination; for thyroid, it assists in analysing ultrasound of adult patients with thyroid nodules suspicious for cancer. The physician selects a region of interest (for thyroid, two orthogonal views of a single nodule); the software returns editable lexicon descriptors and a level-of-suspicion read — for breast, a continuous confidence indicator aligned to the four ACR BI-RADS assessment categories, and for thyroid, a risk read aligned to ACR TI-RADS and American Thyroid Association criteria. The cleared use is assistive throughout: it is not indicated for normal-tissue evaluation, post-surgical sites, or images carrying Doppler or elastography overlays, and it requires physicians to complete prescribed training before use. It performs no triage, no worklist prioritisation and no autonomous read.