U-NET Architecture for Liver Segmentation using Multi Model Scans

Medical Image segmentation performs a very helpful task before diagnosis of a disease as well as pre-surgery. Deep learning models helped a lot in segmentation of medical image and in identifying of tissue characteristics. This research paper is all about designing U-NET architecture for liver tumor segmentation. Present study also reveals the performance analysis of CT scans and MRI scans of different patient collected from different online and offline sources. This research employed U-NET with an encoder-decoder architecture with RELU function as an activation function. In order to get elevated accuracy different operators where experimented. As a result, max-pooling operation was used which gives us maximum accuracy of 98.5%. The data and result verification were done at HCG Hospital, Bengaluru. To measure the segmentation performance Volume Overlapped (VO) (%), Volume Difference (VD) (%), Relative Volume Difference (mm), Average systematic surface Distance (mm), Maximum Surface Distance MSD (mm). This novel architecture has pushed the benchmark of multi model liver scans for segmentation up to 98.5% which is better than the previously reported architectures.