Analysis of 314 Surgical AI Devices Finds Limited Prospective Validation

An analysis published in Annals of Surgery examined the clinical roles, validation evidence, and regulatory pathways of FDA-cleared artificial intelligence and machine learning devices with surgical applications. Researchers systematically screened devices through July 2025, identifying 314 surgically relevant devices, representing 25.1% of those screened. They also assessed predicate networks for differences between cleared devices and their reference devices.

Most devices were imaging-based, with applications in general surgery (24.8%), orthopaedic surgery (18.2%), neurosurgery (17.2%), and cardiac surgery (10.8%). Descriptive and diagnostic functions accounted for 50.6% and 40.4%, respectively; predictive and prescriptive applications accounted for 1.3% and 7.6%. Validation relied mainly on retrospective or bench studies, with only 5.1% supported by higher-quality prospective evidence.

The 510(k) pathway accounted for 95.2% of clearances. Among these devices, 37.5% were classified as having high risk of predicate creep, reflecting substantial differences in intended use or technological characteristics from their predicates. These findings highlight limitations in the evidence supporting clinical adoption, rather than demonstrating patient harm or benefit. The authors call for greater evidence transparency and robust assessment of device equivalence as surgical AI expands toward predictive and interventional roles.