Efficient Region of Interest Detection for Liver Segmentation using 3D CT Scans

Deep learning has become a methodology of choice in medical imaging; one of the applications being classification tasks. The research presented in this paper aims to obtain a region of interest for liver segmentation with the aid of a convolutional neural network to classify 2D slices of a 3D CT volume. This is done by classification of slices to detect slices containing the pelvis and chest so that they can be removed while maintaining the abdomen within which the liver occurs. The presented approach is evaluated on the Medical Image Computing and Computer Assisted Intervention (MICCAI) 2007 grand challenge datasets and the evaluation metrics used are accuracy, recall and precision. The presented approach proved to perform well and the classification models achieved an accuracy rate of 0.99 for pelvis slice classification and 0.97 for chest slice classification.