A New Approach for Liver Plus Its Tumor Segmentation in CT Image by TransNUNet

Computer aided diagnosis (CAD) of human liver and its tumor can provide information for both patients and doctors and help them in disease diagnosis, disease treatment, disease tracking, etc. However, manually done a medical reporting of liver and tumor will cost plenty of time and need expertise to finish the work and reduce the errors. Automatic liver and its tumor segmentation provide efficient way to solve the problem, but there still exist difficulties due to CT image shape, like there is little difference between the healthy part and the diseased part. Our research focuses on computer-only segmentation of liver and its liver tumor in human abdominal CT image. Based on the traditional U-Net network with one attention mechanism, the research has further added one Cbam architecture into the model. We divided the dataset into two parts, first 100 patient samples are used as the training dataset, and last 30 patient samples are used as the valid and testing dataset. (Official dataset provided 130 patient samples in total). We should notice that the dataset not only contains the patient samples that have the tumors, but also samples that do not have. The model was first been trained several times, then validated, and tested with last patient samples using the Liver Tumor Segmentation Challenge (LiTS-2017) dataset. For dice score of liver and its tumor segmentation in training, validated and tested was 98%, 91%, respectively, from the images of LiTS dataset. We provided a solution to deal with the liver and its tumor segmentation problem with computer-only method.