Reading of Small Bowel Capsule Endoscopy After Frame Reduction Using an Artificial Intelligence Algorithm

Abstract

Abstract Objectives Poor visualization images during small bowel capsule endoscopy (SBCE) reading can confuse interpretation of small bowel lesions and increase the physician’s workload. Using a validated artificial intelligence (AI) algorithm that can evaluate the mucosal visualization, we decided to determine whether the SBCE reading after removing poor visualization images could affect the diagnosis. Methods A study was conducted to analyze 90 SBCE cases in which small bowel examination was completed. Two experienced endoscopists performed the frame reduction reading after removing poor visualization images with the AI algorithm (AI user group) and the whole frame reading (AI non-user group). A poor visualization image was defined as image with mucosal visualization < 50%. Outcomes were diagnostic concordance and reading time between the two groups. SBCE diagnosis was classified into Crohn's disease, bleeding, polyp, angiodysplasia, and nonspecific finding. Results The final SBCE diagnosis between the two groups was significantly consistent (k = 0.954, p < 0.001). The mean number of lesion images was 3008.5 ± 9965.0 in the AI non-user group and 1401.7 ± 4811.4 in the AI user group. There were no cases in which lesions were totally removed. Compared to the AI non-user group (120.9 mins), reading time was reduced by 35.6% in the AI-user group (77.9 mins). Conclusions Reading of SBCE after reducing poor visualization frames with AI algorithm did not have a negative effect for final diagnosis. As a reading method integrated with automated lesion detection, frame reduction and mucosal visualization measurement will also be available for comprehensive AI-assisted SBCE interpretation.