Automatic anal sphincter integrity detection from ultrasound images via convolutional neural networks.

This article focuses on the development and evaluation of a deep learning (DL) pipeline using convolutional neural networks (CNNs) for the automatic diagnosis of anal sphincter integrity via pelvic floor ultrasound. The anal sphincter complex, crucial for continence and defecation, is traditionally assessed by experienced sonographers through specific ultrasound planes, a process that is time-consuming and subject to variability. The proposed DL model automates the detection of diagnostic planes, identifies the anal sphincter region, and preliminarily diagnoses sphincter injury, achieving an area under the curve (AUC) of 0.822 and demonstrating strong agreement with expert sonographer assessments. While the study shows promise in improving diagnostic efficiency and accuracy, it is limited by its single-center, relatively small dataset, indicating the need for further validation across diverse clinical settings.