Improved Efficiency and Lesion Detection in Small Bowel Capsule Endoscopy Using the Open-Source Artificial...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Small bowel capsule endoscopy (CE) interpretation is time-consuming and prone to missed lesions.
• An open-source, pretrained artificial intelligence (AI) model, SEE-AI, was evaluated to improve diagnostic performance.

Purpose of the Study:

• To assess if SEE-AI improves diagnostic accuracy and interpretation efficiency in small bowel CE compared to conventional reading.
• To evaluate SEE-AI's performance in suspected small-bowel bleeding (SSBB) cases.

Main Methods:

• Retrospective analysis of 249 PillCam SB3 examinations using a two-reader crossover design.
• SEE-AI generated annotated videos; primary endpoints were lesion detection sensitivity (per-lesion and per-patient).
• Secondary endpoints included specificity, predictive values, accuracy, and reading time; subgroup analysis for SSBB cases.

Main Results:

• AI-assisted reading showed significantly higher sensitivity than conventional reading (per-lesion: 98.8% vs. 86.4%; per-patient: 99.1% vs. 80.3%).
• Mean reading time decreased from 17.9 to 13.7 minutes.
• In SSBB cases, sensitivity for hemorrhagic lesions improved (per-lesion: 98.2% vs. 82.8%; per-patient: 98.6% vs. 73.5%) with reduced reading time.

Conclusions:

• SEE-AI significantly improves lesion detection and reduces reading time for small bowel CE interpretation.
• The AI model maintains openness and reproducibility, potentially reducing clinician workload.
• AI-assisted reading with SEE-AI may become a practical tool and a future standard of care for small bowel CE.