Advanced machine learning for enhanced abdominal organ segmentation

Abstract

This research evaluates the ResUnet model’s performance in using computed tomography (CT) images to segment various abdominal organs. Weak boundaries, computing efficiency, and anatomical diversity are the current obstacles in abdominal multi-organ segmentation. By merging residual networks with U-Net, ResUnet overcomes obstacles by increasing precision and effectiveness, which qualifies it for use in medicine. The model’s effectiveness was assessed on a number of organs, and the segmentation accuracy was measured using the dice similarity coefficient (DSC). The ResUnet model’s ability to precisely segment organs with distinct borders was proved by its exceptional accuracy in segmenting important organs, such as the liver (mean DSC: 0.880) and spleen (mean DSC: 0.830). However, the model struggled to separate the esophagus correctly (mean DSC: 0.000) and struggled with smaller and more complex organs like the pancreas (mean DSC: 0.429) and gallbladder (mean DSC: 0.143). These results highlight the method’s limitations when handling organs with asymmetrical shapes or hazy borders.