An In-Depth Method for Liver Tumor Segmentation and Classification using Trans-UNet and Vision Transformer-based Residual Attention Network

Liver cancers are considered the deadliest tumors with the highest rate of growth globally. Initial liver tumors are categorized as cholangiocarcinoma and hepatocellular carcinoma. A hepatocellular carcinoma is a frequent form of tumor in the liver, which basis for several deaths caused by cancer all over the world. Initial detection of cancers can minimize suffering and boost rates of survival. The implementation of automatic methods that offer accurate segmentation of hepatic cancers is crucial to encourage physicians in the identification of tumors and in preoperative preparation for medical treatment for the liver, which minimizes the risk factors associated with medical operation. The treatment and diagnosis of hepatocellular carcinoma or metastases depend on liver cancer segmentation. Still, effective and automated cancer segmentation is demanding because of vague borders and significant variation within the dimensions, forms, and positions of cancer cells in the liver. Conventional approaches to detecting cancers offer numerous errors during detection, are time-consuming, and require professionals to examine the growth of cancer. In this research work, a liver tumor segmentation and classification framework based on deep learning networks is developed to recognize and cure liver cancer at an early stage. At first, the required images segment and classify the liver tumor are gathered. Then, the segmentation process is carried out via Trans-UNet. The attained segmented images are offered to the liver tumor classification model. This classification is accomplished using the proposed Vision Transformer-based Residual Attention Network (ViT-RAN). The segmentation and classification procedures are helpful to detect the presence of tumors more precisely. The effectiveness of the generated network is verified by validating the outcomes with various deep learning approaches.