Development and Validation of an Artificial Intelligence Model for Small Bowel Capsule Endoscopy Video Review.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Small bowel capsule endoscopy (SBCE) video review is time-consuming for clinicians.
• Developing efficient methods for SBCE analysis is crucial for clinical practice.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) algorithm for automated SBCE video review.
• To assess the performance of the AI tool in detecting CE structured terminology (CEST) findings in real-world clinical settings.

Main Methods:

• A deep learning neural network (SmartScan) was trained on 2927 SBCE examinations from 29 centers to detect 17 CEST findings.
• SmartScan was validated against conventional reading (CR) and SmartScan-assisted reading (SSAR) using 2898 SBCE examinations from 22 centers.

Main Results:

• SSAR detected significantly more CEST findings (95.9%) compared to CR (76.1%).
• The per-patient detection rate for CEST findings was higher with SSAR (79.3%) than CR (70.7%).
• SSAR reduced the number of images requiring review by 96.1% and reading time by 89.3%.

Conclusions:

• A CNN-based algorithm (SmartScan) enhances the detection rate of SBCE findings.
• SmartScan significantly reduces the time required for SBCE video interpretation.
• AI-assisted review holds promise for improving the efficiency and accuracy of SBCE analysis.