A Modified Fuzzy Markov Random Field Incorporating Multiple Features for Liver Tumor Segmentation

Automated segmentation of liver tumors from computerized tomography (CT) images plays a crucial role in computer-aided pathological diagnosis, surgical planning, and postoperative assessment. However, liver tumors exhibit significant variations in size, shape, and location, often low contrast and blurry boundaries with surrounding tissues, making the segmentation task highly challenging. To address this problem, this study proposes a novel and powerful segmentation method based on fuzzy Markov Random Fields(fMRF). Without the need for preprocessing steps, the method utilizes superpixel blocks to form coarse-grained features and intensity, gradient information, and texture information to create enhanced feature vectors representing the tumors. Meanwhile, a new potential function is designed by combining the affiliation distance and multi-feature information in the prior energy function of the model, as a way to further improve the judgment performance of the method for label classification. Finally, morphological processing is applied to refine the segmentation results and obtain the final tumor segmentation outcome. The proposed method is applied to the 3Dircadb dataset. The results demonstrate that this method achieves superior overall segmentation performance compared to many existing approaches, particularly in scenarios involving low contrast and blurry boundaries in tumor segmentation.