Development of an automated ultrasonographic detection method for fecal retention using a transgluteal cleft approach

This study aimed to develop an artificial intelligence-based classification system using ultrasound images obtained via a transgluteal cleft scanning approach for detecting fecal retention in the lower rectum. The goal was to support accurate, objective constipation assessment by nurses in home care settings, where traditional diagnostic tools are often unavailable. Ultrasound videos of the lower rectum were collected from 24 patients undergoing dialysis at a mixed-care hospital. From 90 videos, 2,855 still images were extracted and labeled by expert sonographers based on the presence or absence of hyperechoic areas indicating fecal retention. A deep learning segmentation model using U-Net with a ResNeXt-50 encoder was trained and evaluated. Performance was measured using the intersection over union threshold of 0.5 to define true positives. Accuracy, sensitivity, and specificity were calculated on a test dataset of 758 images. Among the test images, 376 (49.6%) showed fecal retention. The AI system achieved a sensitivity of 81.6%, specificity of 84.0%, and overall accuracy of 82.8%. The mean IoU was 0.601 ± 0.185, indicating a high level of agreement between expert annotations and AI-generated predictions. The tool reliably detected fecal retention in ultrasound images obtained using the transgluteal cleft approach, which overcomes limitations of traditional transabdominal scanning caused by obesity, bladder emptying, or bowel gas. The proposed AI-assisted ultrasound system showed high diagnostic performance in identifying fecal retention in the lower rectum. It may enable non-specialist nurses to assess constipation more safely and accurately, particularly in home-care environments. This technology has the potential to reduce unnecessary laxative use and invasive interventions, ultimately improving the quality of bowel care for older adults with impaired communication or mobility.