Development and clinical assessment of a novel AI-based diagnostic model for Hirschsprung's disease.

This study aimed to develop an AI-based diagnostic model for Hirschsprung's disease (HD) using deep learning on contrast enema (CE) images, with the goal of improving diagnostic accuracy while reducing invasiveness. The dataset included 725 CE images from histopathologically confirmed HD patients from 2013 to 2022. Employing Python and PyTorch, a deep learning model based on the YOLOv8 algorithm was trained and validated, emphasizing key metrics like mean average precision (mAP), precision, recall, and F1 score. This model exhibited high precision (0.87477) and recall (0.87317), with an mAP50 score of 0.91. External validation showed promising results, including a sensitivity of 86.96%, a specificity of 72.22%, and an overall accuracy of 80.49%. This AI model offers a less-invasive and accurate alternative to traditional HD diagnostics, especially beneficial for initial screening in pediatric gastroenterology, with the potential to enhance healthcare diagnostics through AI integration.