AI Improves Lesion Detection in IBD Over Standard Methods

BERLIN, Germany — Artificial intelligence (AI)–assisted capsule endoscopy (CE) readings showed higher sensitivity and accuracy in detecting ulcers and erosions in patients with inflammatory bowel disease (IBD) than did conventional readings in a first-of-its-kind, multicenter study. 

In addition to the model’s superior diagnostic performance than standard of care, it also achieved a significant reduction in the mean reading time per exam. 

Furthermore, the study clinically validated an AI model in real time for small-bowel CE. 

The AI model addresses long-standing limitations of CE interpretation, including time-consuming readings and inter-observer variability.

“It’s a huge improvement on the technology readiness level of the AI model,” said senior study investigator Miguel Mascarenhas, MD, PhD, Head of Precision Medicine Unit Hospital São João, Faculty of Medicine of the University of Porto, Porto, Portugal. 

Until now, there has been no AI system using a CE platform that has proven so effective in so many real-life clinical settings, he explained. “This technology is set to transform endoscopic practice and clinical management in inflammatory bowel disease.” 

The findings (abstract DOP089) were presented at European Crohn’s and Colitis Organisation 2025 Congress by Francisco Mendes, MD, a resident in gastroenterology, also at the Hospital São João. 

More Lesions, Less Time

Researchers conducted the prospective study involving centers in Portugal, Spain, and the United States between January 2021 and April 2024. Two CE devices (PillCamSB3 and Olympus EC-10) were analyzed for their performance across 137 CE exams in 137 patients, 49 of whom had Crohn’s disease. AI-assisted readings were compared with standard-of-care readings, with expert board consensus considered to be the gold standard. Key performance metrics included sensitivity, specificity, positive predictive value (PPV), and negative PV (NPV). 

During expert board review, ulcers and erosions were identified in 56 patients (40.9%), with a sensitivity of 60.7%, specificity of 98.8%, a PPV of 97.1%, and an NPV of 78.4%, leading to an overall accuracy for the detection of ulcers and erosions of 83.2%.

In comparison, the AI-assisted readings outperformed conventional readings with a sensitivity of 94.6%, specificity of 80.2%, a PPV of 76.8%, an NPV of 95.6%, leading to an overall accuracy of 86.1%.

The AI-assisted model diagnosis was noninferior (P P 

In addition, the mean time taken per reading was under 4 minutes (239 seconds) per exam for AI compared with around 1.0-1.5 hours for standard-of-care readings.

The increased diagnostic accuracy of this AI model done in far less time allows us to engage more with the patient and attend to other care-related tasks, Mascarenhas said. 

CE has great potential not only in IBD but also in other GI-related screening, including colorectal cancer screening, he added. Once the bottleneck of reading time with CE is solved, it will become the first-line tool for screening.

Reading time is “one of several barriers” to integration of CE into clinical practice, Shomron Ben-Horin, MD, Director, Sheba Medical Center, Tel-Aviv University, Tel-Aviv, Israel, told Medscape Medical News. But it “is the most accurate modality for detection of inflammatory activity along the entire small bowel.”

Based on these study results, AI is the way to go, said Ben-Horin, who was not involved in the study. “There was even a signal for better accuracy, which is intriguing,” he added. This study points toward AI being more accurate than the physicians in reading, and that is important.

Also commenting was Miles Parkes, MD, consultant gastroenterologist at Addenbrooke’s Hospital in Cambridge, United Kingdom. 

“Both the sensitivity and the specificity of the output are reassuring, but there might be some devil in the detail, he told Medscape Medical News. However, as a general principle the performance of this model is impressive.” 

Mascarenhas and Mendes declared no financial disclosures. Ben Horin received fees from Medtronic to attend the conference. Parkes declared no financial disclosures.