Liver Tumor Segmentation using Resnet based Mask-R-CNN

Liver tumor segmentation is the most significant and challenging task due to high variation in size, location, depth of tumor and intimate boundaries of other organs around the liver. Computed Tomography (CT) scans of the liver differ in size and other parameters, including the number of layers in each CT scan. Thus, liver CT scans from different parameters make it a challenging task. Thus, we propose an efficient liver tumor segmentation method with Mask Region-Convolutional Neural Network (Mask-RCNN) with resnet 101 as the backboneto come upwith these challenges. The proposed method based on the Mask-RCNN approach effectively identifies tumor in liver through generating proposals about the tumor region. The proposed method converts CT volumes to normalized CT image slices in preprocessing step to get a cardinal outline of liver. Then normalized CT images are feed to resnet 101 for feature extraction. Then finally, Mask-RCNN is appliedto segment tumors in the liver. Our experiment dataset comprises 130 CT scans of different patients taken from different clinical sites, and it is freely available atLiTS challenge website. The proposed method is trained on converted CT image slices. Our proposed method hasimproved the segmentation rate by providing segmentation accuracy with average dice coefficient, JC, VOE and RVD of 0.95, 0.41, 0.12, and 0.15, respectively and lesser training time due to converted CT image slices instead of training on CT volumes.