Liver Tumour Segmentation based on ResNet Technique

It is known that the sixth most common cancer worldwide is liver cancer and CT scans are commonly used to diagnose liver cancer. Hence in this study, deep learning techniques specifically the ResNet models are used to extract the liver and tumour from the CT scans. Here, four liver segmentation methods are used based on 130 CT datasets namely the ResNet-18, ResNet-34, ResNet-50, and ResNet-101. Each model is evaluated and validated based on their training and testing accuracy, number of epochs, valid loss and train loss. Initial results showed that the highest accuracy is contributed by ResNet-34 with 99.2% accuracy and next is ResNet-50. Additionally, ResNet-101 is the most efficient network model whilst ResNet-18 is the most rapid. These findings proved that the deep learning can be used for segmentation of liver tumour based on the CT scan images.