Automated Classification of Liver Cancer Stages Using Deep Learning on Histopathological Images

Abstract

Plagued by a high incidence rate worldwide, liver cancer comes in sixth place among all cancers. The degree of differentiation, which may be roughly divided into three types: weakly moderately differentiated, highly differentiated, and differentiated, has a substantial impact on the malignancy level of this terrible illness. In direction to improve the existence of affected role and life expectancy, therapeutic techniques that are customised for these varied levels of diversity are essential. The gold standard for identifying the main liver cancer, hepatocellular carcinoma (HCC), is a histopathological picture that allows for exact distinction of liver tumours at different stages of development. This study explores the creative use of the R-CNN algorithm for deep learning for the astute categorization of histological pictures associated with liver cancer that undergoes differentiation. So this study compared how well R-CNN did compared to five other popular deep learning models -SKNet, ResNet CBAM, ResNet50, VGG16, and SENet. It was really important to set up a good system to collect a lot of different data for this project. This would make sure they had enough info to properly test how well the different math models worked. They also created a thorough and precise method using things like recall, confusion matrices, F1-scores, and accuracy to analyze how the models performed. The results showed that R-CNN did amazingly well, with an accuracy of 96.7%! That means it was able to classify things correctly almost all the time. This demonstrates it had the most accurate predictions out of everything they looked at. Additionally, the R-CNN model proved to be very reliable and able to generalize well. In other words, it should work just as good on new, unseen data as it did on the information it was originally skilled on. This study compares the routine of R-CNN to five other well-established deep learning models: SKNet, ResNet CBAM, ResNet50, VGG16, and SENet. Developing a robust data collection infrastructure is critical to enable the project to ensure a large and varied dataset for thorough evaluation of the mathematical models under consideration. A comprehensive and precise evaluation approach was provided through judicious usage of metrics like F1-Score, recall, confusion matrix, and accuracy to analyze the model performance. Testing results demonstrate R-CNN's capabilities, evidenced by its notable 96. 7% accuracy indicating highly precise classification outcomes. Furthermore, the model exhibits strong reliability and generalizability.