Automatic Liver Tumor Segmentation Based on Random Forest and Fuzzy Clustering

This paper presents an automatic method for liver tumor segmentation in CT scans. The proposed segmentation algorithm is a two-stage process. In the first step, the curvature filter is employed for removing the noise in CT images, and a trained mask is used to be a spatial regularization to constrain our segmentation in a specific region. In the second step, basing on the preprocessed results, we combine random forest with fuzzy clustering to segment liver tumor. In the experiments, the proposed method obtains promising results on the liver tumor segmentation challenge testing dataset. The calculated mean scores of Dice, volume of overlap error (VOE), relative volume difference (RVD), average symmetric surface distance (ASD), and maximum symmetric surface distance (MSD) are 0.47, 0.65, −0.35, 11.49, and 64.31, respectively.