Enhanced U-Net for Computer-aided Diagnosis with Cetacean Postmortem CT Scan

Postmortem computed tomography (PMCT) scan has long been used in the postmortem examination of cetaceans on Virtopsy to find out the cause of death such as diseases on internal organs. However, manual diagnosis using PMCT scans is labour-intensive and time consuming. Driven by the recent advances in deep learning, in this project, we develop a computer-aided diagnosis system based on U-Net, a convolutional neural network, for cetaceans. In addition, we propose two loss functions to resolve the bias due to the skewed distribution of the different areas in the PMCT-scans, thereby improving the system accuracy significantly.