Automated liver image segmentation using entropy-based thresholding and median filtering

The separation of liver images from abdominal scans has emerged as a critical focus in biomedical image processing, serving as a foundational step in automated techniques for liver disease diagnosis, treatment planning, and follow-up assessment. Current medical research and case studies underscore the challenges of liver segmentation, primarily due to the low contrast between the liver and surrounding tissues in computed tomography (CT) images. Furthermore, the liver's edges are often indistinct, and its texture, shape, color, and size exhibit significant variability. With advancements in medical imaging technology, the volume of data requiring processing has grown substantially, highlighting the need for automated methods to replace time-intensive manual segmentation procedures. In response to these challenges, a novel threshold-based segmentation technique has been introduced, utilizing liver image entropy as a measure of information content. The process involves denoising with a median filter, followed by cropping a random section of the liver image to determine its entropy distribution. This distribution establishes upper and lower bounds, facilitating precise separation of the liver from its background. The proposed method was evaluated on CT scan images from 60 patients, addressing diverse and complex segmentation scenarios. Key performance metrics, including maximum edge distance (MED), relative volume difference (RVD), accuracy, and dice similarity factor (DSF), were employed to benchmark the model against expert-traced reference images. The results indicate an average MED of 12.5 mm, an average RVD of 4.2%, an average accuracy of 91.70%, and an average DSF of 90.95%. These results demonstrate the effectiveness of the proposed model as a robust tool for computer-aided decision support systems, significantly advancing the accuracy and reliability of clinical diagnosis. © 2024 Sangeeta K Siri et al.