Segmenting Liver Volume for Surgical Analysis

Introduction: Almost two million people worldwide die annually due to hepatic-related diseases. Half of these diseases are attributed to cirrhosis and the other half are related to hepatitis and hepatocellular carcinoma (HCC). The liver is also a metastasis hub from adjacent organs. This research aims to create an accurate high-quality delineation of the human liver and prepare them to be 3D printed for medical analysis to help aid medical practitioners in pre-procedural planning. Materials and Methods: Convolutional neural networks (ConvNets) are used to perform the liver tissues delineation. A famous ConvNet, named U-net, is used as the basis benchmark architecture that is also known for its great outcomes in the medical segmentation field. Contrast-enhanced computerized tomography (CT) scans are used from the famous Medical Segmentation Decathlon Challenge (Task 8: Hepatic Vessel), abbreviated as MSDC-T8. It contains 443 CT scans, which is considered the largest dataset that contains both the tumors and vessels ground-truth segmentation. Some researchers also generated the liver masks for this dataset, making it a complete dataset that contains all the relevant tissues’ ground-truth masks. Results: Currently, the liver delineation has been successfully done with very high DICE = 98.12% (higher than the state-of-the-art results DICE = 97.61%), where a comparison between two famous schedulers namely, ReduceLRonPlateau and OneCycleLR has been conducted. Moreover, the 3D liver volume creation has also been completed and built via the marching cube algorithm. Conclusions/Future Directions: The developed ConvNet can segment livers with high confidence. The tumor(s) and vessels tissues segmentation are also under investigation now. Moreover, newly devised self-organized neural networks (Self-ONN) look promising and will be investigated soon. Lastly, a GUI will be built so that the medical practitioner can just insert the CT volume and get the 3D liver volume with all the segmented tissues.