Physician in the Loop
For youSign in
General surgery ·

Review finds little prospective evidence behind surgical AI devices; robotic Whipple training program reports results; AI model maps hernia anatomy

An instrument tray lined with a blue towel, holding a scalpel with an amber handle, toothed forceps and curved scissors laid side by side.
THE SHORT READ

Only 5.1% of 314 surgical AI devices cleared by the Food and Drug Administration (FDA) are backed by higher-quality prospective evidence, an Imperial College London analysis in Annals of Surgery found. A pan-European program reported outcomes from 486 robotic pancreatoduodenectomies at 20 centers new to the operation, and researchers described an AI model that recognizes key anatomy in hernia repair video.

THE DETAILS
  • AI Oct 8, 2026
    Analysis of 314 FDA-cleared surgical AI devices finds 5.1% backed by higher-quality prospective evidence
    Researchers at Imperial College London screened Food and Drug Administration (FDA)-cleared AI and machine learning devices through July 2025 and found 314 (25.1%) with surgical applications, according to a study in Annals of Surgery. General surgery accounted for 24.8%, and most devices were descriptive or diagnostic imaging tools. Only 5.1% were supported by higher-quality prospective evidence, and the 510(k) pathway accounted for 95.2% of clearances. Among those, 37.5% showed a high risk of predicate creep, meaning substantial differences in intended use or technology from their predicate devices, the authors said.
    Sources: Annals of Surgery, PubMed, Crossref
  • Technology Oct 7, 2026
    LEARNBOT program trains 20 European centers in robotic pancreatoduodenectomy, with 18.3% conversion across 486 cases
    A structured training program for robotic pancreatoduodenectomy at 20 European high-volume centers in 12 countries, none with previous experience in the operation, combined a video library, biotissue simulation, off-site case observation and on-site proctoring, according to a prospective multicenter study in Annals of Surgery. Across 486 procedures, the conversion rate was 18.3%, Clavien-Dindo grade III or higher complications occurred in 31.4%, pancreatic fistula in 21.2%, and in-hospital or 30-day mortality was 2.7%. The authors said a learning-curve effect in major complications remained detectable per center.
    Sources: Annals of Surgery, PubMed, Crossref
  • AI Oct 8, 2026
    AI model recognizes six anatomic structures in totally extraperitoneal hernia repair video at 60 frames per second
    Researchers trained an object detection model on frames from 30 videos of totally extraperitoneal (TEP) inguinal hernia repair to recognize six key anatomic structures, including the triangle of pain and the triangle of doom, according to a study in Surgical Endoscopy. On held-out frames, the model achieved precision of 0.805 and recall of 0.704, with inference at 60 frames per second. The model was evaluated on annotated still frames, not during live operations, and the study did not measure patient outcomes.
    Sources: Surgical Endoscopy, Springer, PubMed

The weekly innovation roundup for general surgery, published every Sunday for the week before; every item links to its sources. Latest general surgery news · All specialties