Loading ...

TOPLINE:

An artificial intelligence (AI)–assisted capsule endoscopy reading may provide more accurate and faster detection of small bowel bleeding lesions than a standard reading, a multicenter prospective trial suggested.

METHODOLOGY:

• Patients with suspected small bowel bleeding were prospectively enrolled at 14 European centers.
• Participants underwent small bowel capsule endoscopy with the NaviCam SB system (Ankon, China), which includes a deep neural network-based AI system (ProScan) for automatic detection of lesions.
• An initial reading was done in the standard reading mode, whereas a second blinded reading was performed with AI assistance; that is, the AI ran an automated reading first, and only AI-selected images were assessed by human readers.
• The primary endpoint was to assess the noninferiority of the AI-assisted reading vs the standard reading in the detection (diagnostic yield) of potential small bowel bleeding P1 and P2 lesions (according to the adapted Saurin classification) in a per-patient analysis.

TAKEAWAY:

• A total of 133 patients were included in the final analysis (mean age, 66.5 years; 55% women).
• The per-patient analysis showed that the diagnostic yield of the P1 and P2 lesions in the AI-assisted reading (98 lesions; 73.7%) was noninferior and superior to the standard reading (82 lesions; 62.4%).
• The miss rate of standard readers (21.0%) was significantly higher than that of AI-assisted readers (6.6%). The average time to detect a lesion was 18.7 minutes for standard readers and 1.6 minutes for AI-assisted readers, and the average number of lesions detected per patient was 1.8 by standard readers and 2.4 by AI-assisted readers.
• The mean small bowel reading time was 33.7 minutes in the standard reading mode and 3.8 minutes in the AI-assisted reading mode.

IN PRACTICE:

"Our results provide novel and specific evidence of both the noninferiority and superiority of capsule endoscopy AI-assisted reading versus standard reading in detection of specific types of lesions in patients with capsule endoscopy," the authors wrote. "In addition, our results confirm the impact of the use of AI in the remarkable reduction of the reading time."

SOURCE:

The study, led by Cristiano Spada, MD, PhD, Fondazione Poliambulanza Istituto Ospedaliero, Brescia, Italy, was published online in the May issue of The Lancet Digital Health.

LIMITATIONS:

The study had several limitations, including no assessment of inter-reader variability, a relatively small sample size, no evaluation of clinical outcomes, and a focus on only patients with suspected small bowel bleeding.

DISCLOSURES:

Ankon Technologies and AnX Robotica provided the NaviCam SB system. Spada received support for attending meetings and travel from AnX Robotica and honoraria from AnX Robotica for the adjudication committee. Several other coauthors also received fees from AnX Robotica.