Colon Cancer Detection Using A Lightweight-CNN With Grad-CAM++ Visualization

Colon cancer is caused by cells that proliferate in the colon, a part of the large intestine. Early detection of cancer may make treatment easier and save many lives each year. The goal of this study is to identify improved approaches to diagnosing colon cancer rapidly and correctly. It intends to accomplish this through the use of a computer program known as a lightweight and efficient Convolutional Neural Network (CNN), which operates quickly and requires little computational power. The method may aid doctors in detecting colon cancer more successfully than earlier. The proposed model outperforms prior efforts with an amazing validation accuracy of $99.48 %$. In our study, we employed 10,000 images from two separate classes to distinguish colon cancer cells from normal colon cells. The suggested model performs admirably, as indicated by its remarkable precision, recall, and F1-score, all of which are greater than $99 %$, as well as its perfect Area Under Curve (AUC) score of $100 %$. The parameters in the suggested model are lower than those in the current transfer learning models. We also compared our novel CNN model to the industry-recognized MobileNetv2 transfer learning standard to demonstrate our method's remarkable effectiveness. To illustrate the model's decision-making process, Gradient-weighted class activation mapping, or Grad Cam++, has been used.