Colon cancer detection using VGG16 and XAI methods

Colon cancer is responsible for a high percentage of cancer mortality throughout the world, and therefore there was a need to create early and better diagnosis methods. The conventional methods were not unambiguous enough and were “black boxes,” which compromised clinical adoption and endorsement. In this paper, we attempt to address these issues by creating a VGG16-tailored convolutional neural network on a balanced dataset of 10,000 histopathological images of 5,000 colon adenocarcinomas and 5,000 benign tissue samples. We used the latest breakthroughs in deep learning in medical imaging to include XAI techniques which are Grad-CAM, LIME and SHAP to provide both visual and quantitative explanations of model decisions. This illuminates the attributes behind the classification, which are concealed usually. Our work highlights the efficiency of combining high-performing CNN architectures with explainable artificial intelligence to enhance the early detection of colon cancer and facilitate clinical diagnosis with increased transparency and reliability.