Multimodal AI-Assisted Bowel Preparation for Colonoscopy

Overview

This randomized controlled trial aims to evaluate the effectiveness of a multimodal artificial intelligence (AI)-assisted smartphone application in improving bowel preparation outcomes among hospitalized adults undergoing colonoscopy. A total of 140 participants will be randomly assigned in a 1:1 ratio to either an experimental group or a control group. The control group will receive conventional written and verbal nursing education, while the experimental group will receive the same standard education plus a multimodal AI-assisted application delivered through the LINE platform. The application provides structured bowel preparation education, interactive AI-based question-and-answer support, dietary image recognition, stool image analysis, and individualized feedback. Study outcomes will include bowel preparation knowledge, satisfaction with nursing education, and bowel cleansing quality assessed using the Aronchick Scale.

Adequate bowel preparation is essential for high-quality colonoscopy because insufficient bowel cleansing may reduce mucosal visualization and affect the detection of colorectal lesions. Conventional bowel preparation education generally relies on written materials and verbal instructions. However, patients may have difficulty understanding dietary restrictions, bowel cleansing medication instructions, and whether bowel cleansing is adequate during the preparation process. This study evaluates a multimodal AI-assisted smartphone application designed to support patients throughout the bowel preparation process. The application is integrated with the LINE platform and provides structured educational content, including bowel preparation instructions, bowel cleansing medication guidance, dietary preparation, instructional videos, and an interactive AI chatbot for real-time individualized responses. The application also incorporates two AI-assisted image-recognition functions. First, participants may submit photographs of their meals before colonoscopy. A multimodal AI model analyzes the food images and classifies the dietary pattern as a clear liquid diet, low-residue diet, or regular diet, and provides individualized dietary recommendations. Second, after taking bowel cleansing medication, participants may submit stool images. The system uses a multimodal AI model in combination with a convolutional neural network model to assess bowel cleansing status and provide feedback. Participants in the control group will receive conventional written and verbal nursing education regarding diet, bowel cleansing medication, and colonoscopy preparation. Participants in the experimental group will receive the same conventional education plus access to the multimodal AI-assisted application. Bowel preparation knowledge will be assessed before and after the intervention, satisfaction with nursing education will be evaluated after colonoscopy, and bowel cleansing quality will be assessed by the endoscopist using the Aronchick Scale.