Automated Aortic Stenosis Software (AutoAS)

K254161

GE Medical Systems Ultrasound & Primary Care Diagnostics, LLC · cleared 2026-03-27 · product code POK · Radiology

Premarket evidence — what FDA accepted

Device typesamd
source quote (p.4)
“AutoAS is a software application intended to assist medical professionals in the assessment of moderate/severe aortic stenosis (AS). The software uses an artificial intelligence (AI) algorithm to process previously acquired two-dimensional transthoracic echocardiography (2D-TTE) images to provide a suggestion of moderate/severe aortic stenosis along with an associated confidence metric that can be a diagnostic aid to a physician in a point of care or similar setting in determining if further evaluation is needed, including whether a full echocardiogram (2D, Doppler) needs to be performed. AutoAS software is indicated for use in adult patients and is intended to be an accessory to compatible ultrasound systems in environments where healthcare is provided.”
Algorithmdeep-learning artificial intelligence
source quote (p.12)
“Both devices utilize deep-learning artificial intelligence as the core technology to provide diagnostic aid to the user in the assessment of heart conditions.”
Adaptive (vs locked)FDA source did not state this
PCCPFDA source did not state this
Cybersecurity addressedYes
source quote (p.7)
“Software documentation generated as part of the design process included: Cybersecurity”

Validation studies (5)

Bench

sample size not stated

endpoints: accurately detecting whether moderate / severe aortic stenosis was present

Bench

sample size not stated

endpoints: positive predictive value (PPV); Sensitivity

Bench

sample size not stated

endpoints: statistically significantly lesser MAD / MAE than the established benchmark

Retrospective clinical

n=401 patients · 4 site(s)

endpoints: Area Under the ROC Curve; Specificity; Sensitivity

standards: American Society of Echocardiography (ASE) clinical guidelines

Reader study (MRMC)

n=220 patients · 3 site(s)

endpoints: diagnostic performance; sensitivity; specificity; partial AUROC; inter-rater agreement

Reported performance (13 observations)

sensitivity0.752CI 67.4% - 83.0%
source quote (p.9)
“Sensitivity of 75.2% [95% CI: 67.4% - 83.0%]”
specificity0.924CI 86.3% - 98.4%
source quote (p.9)
“Specificity of 92.4% [95% CI: 86.3% - 98.4%]”
aurocas written: “auc”0.932CI 90.5% - 95.6%
source quote (p.9)
“Area Under the ROC Curve, 93.2% [95% CI: 90.5% - 95.6%]”
ppvas written: “PPV (Clip Annotator)”1CI (98.5%, 100.0%)
source quote (p.8)
“Testing demonstrated both a positive predictive value (PPV) and Sensitivity of 100% (95% CI: (98.5%, 100.0%)) across all view types (i.e., PLAX, AP5, PSAX-AV, and all other views) when classifying the B-mode image.”
sensitivityas written: “Sensitivity (Clip Annotator)”1CI (98.5%, 100.0%)
source quote (p.8)
“Testing demonstrated both a positive predictive value (PPV) and Sensitivity of 100% (95% CI: (98.5%, 100.0%)) across all view types (i.e., PLAX, AP5, PSAX-AV, and all other views) when classifying the B-mode image.”
ppvas written: “PPV (Clip Annotator - view classification)”0.971CI (94.2%, 98.8%)
source quote (p.8)
“for any image that was classified as B-mode, the ability to accurately classify the view was also tested, and the verification test results revealed a PPV of at least 97.1% (95% CI: (94.2%, 98.8%))”
sensitivityas written: “Sensitivity (Clip Annotator - view classification)”0.875CI (83.1%, 91.2%)
source quote (p.8)
“and a Sensitivity of at least 87.5% (95% CI: (83.1%, 91.2%)) across all view types.”
sensitivityas written: “Sensitivity improvement (Reader Study)”0.055CI (1.5%, 9.5%)
source quote (p.11)
“A statistically significant improvement in sensitivity was observed for the “Aided” readers compared to the "Unaided" readers (+ 5.5%, 95% CI: (1.5%, 9.5%))”
specificityas written: “Specificity (Reader Study - Aided)”0.897
source quote (p.11)
“while maintaining comparable specificity (0.897 vs. 0.900).”
specificityas written: “Specificity (Reader Study - Unaided)”0.9
source quote (p.11)
“while maintaining comparable specificity (0.897 vs. 0.900).”
aurocas written: “Partial AUROC difference (Reader Study)”0.089CI 1.2%, 20.5%
source quote (p.11)
“Furthermore, when comparing the diagnostic performance of the two reader groups, the critical region of the ROC curve revealed superiority for the “Aided” group with an 8.9% [95% CI: 1.2%, 20.5%] difference in partial AUROC.”
agreement_kappaas written: “Inter-rater agreement (Reader Study - Aided)”0.89
source quote (p.11)
“In addition, aided readers demonstrated higher inter-rater agreement (89.0%) than unaided readers (81.9%), comparable to the reference standard (88.7%), reflecting improved reader consistency and diagnostic performance.”
agreement_kappaas written: “Inter-rater agreement (Reader Study - Unaided)”0.819
source quote (p.11)
“In addition, aided readers demonstrated higher inter-rater agreement (89.0%) than unaided readers (81.9%), comparable to the reference standard (88.7%), reflecting improved reader consistency and diagnostic performance.”

Each value carries its own analysis unit and task — never compare or pool across devices. Source: 510(k) summary PDF.

Predicate network

Postmarket — what happened after clearance

0
recalls in product code, 24mo
0
MAUDE reports in code, 12mo
—
vs code's own 3-yr baseline
0
drift signals on this device

Recall and MAUDE counts are product-code-level (reports aren't reliably attributable to one device); a recall is shown as device-attributed only when the recall record itself lists this clearance number. Signals are descriptive observables with sources — never a judgment that the device is unsafe or drifting. Snapshot 2026-07-08.

Reimbursement — how devices like this got paid

Not yet tracked — no payment pathway indexed for this clearance (the reimbursement corpus is a growing seed set).

Applicable FDA guidance — what the submission is measured against

FDA guidance documents and guiding principles applicable to 510(k) AI/ML devices in the Radiology panel. A curated reference index, not legal or regulatory advice — each item states its own status, and a draft is never binding.

Applicability is derived from the device's FDA advisory panel and pathway — cross-cutting guidances apply to every AI/ML device; panel-specific ones are flagged. Titles, dates, and links verified against fda.gov as of July 2026.

Constat Precedent · public FDA/CMS data · descriptive decision-support, not regulatory or reimbursement advice. Share this page: constat.dev/precedent/device/K254161