Enhanced Diagnosis of Liver Cancer Subtypes Through Deep Learning Techniques

Abstract

Liver cancer presents a substantial threat to life and is recognized as one of the rapidly growing forms of cancer globally. According to a 2020 World Health Organization study, Egypt ranks second internationally in liver cancer mortality, accounting for 4.57% of total fatalities. Liver cancer may be categorized into several categories, with the detection and treatment of each stage of differentiation being critical in deciding patients' survival rate and duration. In the last few years, the invention of a computer-assisted imaging procedure that employs deep learning has become critical for evaluating liver illnesses, consequently improving medical diagnosis. This paper makes several contributions. First, it presents a significant effort in creating a multi-class liver tumor dataset sourced from Ain Shams University Specialized Hospital (ASUSH) in Egypt and annotated by two specialists. The dataset comprises a total of 8237 CT images categorized into eight distinct classes. Furthermore, the paper benchmarks the dataset to multiple deep learning models, including a proposed CNN, as well as fine-tuned pre-trained models like VGG16, VGG19, ResNet50, ResNet101, Xception, and Inception-V3. The results of the experiment demonstrate that the suggested CNN surpasses the fine-tuned networks, obtaining an average accuracy of 99.76% in predicting eight-class liver cancer variations.