Automatic Tumor Segmentation Using Machine Learning Classifiers

Abstract

Segmentation of liver and tumor from abdominal Computed Tomography (CT) is important for proper planning and treatment of liver disease. Variable size, intensity overlap, and complexity of CT images probe a problem for a radiologist. These issues make accurate and reliable delineation of liver and tumor very difficult and time-consuming. So, an automatic method is desired and beneficial. In this paper, we propose a fully automatic method to segment both liver and tumor using an array of Gabor Filter (Gabor Bank(GB)) and Machine Learning (ML) classifiers: Random Forest (RF) and Deep Neural Network (DNN). First, GB extract pixel level Gabor features from CT images. Secondly, the liver is segmented using ML classifiers trained on Gabor features. Finally, tumor segmentation is done on the segmented liver image using the same approach as in liver segmentation. 31 CT image slices containing hepatic tumors from 3D-IRCADb (3D Image Reconstruction for Comparison of Algorithm Database) were used to validate our proposed method. For liver segmentation, the experimental result showed that the proposed method with RF classifier performed better than DNN, and can achieve high performance of 99.55% accuracy and 99.03% dice similarity coefficient. Also, for tumor segmentation, a similar conclusion was drawn.