Soft optimization techniques for automatic liver cancer detection in abdominal liver images

Abstract

Automatically segmenting the liver is a challenging process, and segmenting the tumour from the liver adds another layer of complexity. Because of the overlap in intensity and fluctuation in location and form of soft tissues, segmenting the liver and tumour from abdominal Computed Tomography (CT) images merely based on grey levels or shape is very undesirable. To address these challenges, this study proposes employing Gabor Features (GF) and three distinct machine learning algorithms: Random Forest (RF), Support Vector Machine (SVM), and Deep Neural Net, a more efficient way of liver and tumour segmentation from CT images (DNN). The texture data produced by GF should be consistent and homogeneous across numerous slices of the same organ. In the first, pixel level features are extracted using an array of Gabor filters. Second, utilising three separate classifiers: RF, SVM, and DNN trained on GF, liver segmentation is conducted to remove liver from an abdominal CT picture. Finally, using GF and the same set of classifiers, tumour segmentation is performed on the segmented liver image.