Physician in the Loop
For youSign inAsk
Gastroenterology ·

AI more than doubles capsule endoscopy sensitivity; chatbots miss GI emergencies more in Vietnamese than English; TissueCypher and PIB disagree

A flexible endoscope with its control head and two round angulation wheels, the long insertion tube coiling in a loose loop to a glowing amber tip.
THE SHORT READ

AI-assisted capsule endoscopy reading raised sensitivity for protuberant lesions to 93.0% from 44.0% across 13 international centers, according to a featured abstract at the American College of Gastroenterology (ACG) 2026 meeting.

Another ACG abstract found large language models failed to recommend emergency care more often in Vietnamese than in English, and Castle Biosciences reported slight agreement between its TissueCypher test and a clinicopathologic score in Barrett's esophagus.

THE DETAILS
  • AI Oct 11, 2026
    AI-assisted capsule reading raises sensitivity for protuberant lesions to 93.0% from 44.0% in 13-center study
    In a real-world, multidevice study across 13 international centers presented at the American College of Gastroenterology (ACG) 2026 meeting, AI-assisted reading of capsule endoscopy videos more than doubled diagnostic sensitivity for protuberant lesions compared with conventional reading, 93.0% versus 44.0%. The AI-assisted approach had a negative predictive value of 98.3% and a median reading time of less than 6 minutes per video, according to ACG. ACG research was embargoed until Oct. 11.
    Source: ACG
  • AI Oct 11, 2026
    Language models fail to recommend emergency care in 15% of English and about 41% of Vietnamese GI scenarios
    A multilingual safety and accuracy benchmark presented at the American College of Gastroenterology (ACG) 2026 meeting tested four large language models across the seven most common U.S. languages on gastroenterology scenarios. Failure to recommend emergency care rose from 15% in English to approximately 41% in Vietnamese, according to ACG's summary of the abstract. ACG's summary did not name the models.
    Source: ACG
  • Technology Oct 12, 2026
    TissueCypher and Progression in Barrett's score show slight agreement on Barrett's esophagus risk, Castle says
    Castle Biosciences said Oct. 12 that a multicenter study at the American College of Gastroenterology (ACG) 2026 meeting compared its TissueCypher test, which uses an AI-derived spatialomics approach on routine endoscopic biopsies, with the Progression in Barrett's (PIB) clinicopathologic score. Agreement was slight (Cohen's kappa 0.04): TissueCypher classified 83% of patients as low risk and 17% as intermediate or high risk, versus 31% and 69% with PIB. Management recommendations changed in line with the test's risk class for 96% of patients, the company said.
    Source: Castle Biosciences

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