A neural network-based algorithm for assessing the cleanliness of small bowel during capsule endoscopy

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Small-bowel capsule endoscopy (SBCE) cleanliness scoring lacks reproducibility.
• Automated assessment is needed to standardize SBCE interpretation.

Purpose of the Study:

• To evaluate a neural network algorithm for automated small-bowel cleanliness assessment in SBCE.
• To determine the algorithm's accuracy and reproducibility compared to expert readers.

Main Methods:

• A database of 600 SBCE still frames was used for training.
• 156 SBCE videos were categorized by experts for testing the algorithm.
• The algorithm was tuned and evaluated on independent video subsets.

Main Results:

• The algorithm achieved 90.3% sensitivity, 83.3% specificity, and 89.7% accuracy.
• Perfect reproducibility was observed.
• Mean calculation time per video was 3 minutes.

Conclusions:

• The neural network algorithm provides a highly sensitive and automated method for assessing small-bowel cleanliness.
• This technology supports automated and standardized SBCE reporting.