Enhanced Colon Cancer Screening with Endoscopic Images Using Deep Learning Techniques

Colon cancer is a leading cause of cancer- related deaths worldwide, where early detection is vital for improving outcomes. This study focuses on a deep-learning approach for detecting colon cancer using endoscopy images. By employing four pre-trained models—EfficientNetB2, InceptionV3, MobileNetV3, and ResNet50—images are classified into normal, ulcerative colitis, polyps, and esophagitis categories. Public datasets like KVASIR and ETIS-Larib-Polyp DB are utilized, with the original 800 images expanded to 3,200 through data augmentation to enhance model generalization. Performance is measured using metrics such as accuracy, precision, recall, and F1-score. The results indicate that the accuracies of InceptionV3, EfficientNetB2, ResNet50, and MobileNetV3 are 94%, 93%, 89.12%, and 87%, respectively. Given its strong accuracy and resource efficiency, InceptionV3 emerges as the best-performing model, making it suitable for deployment in low-resource settings. This study highlights the potential of deep learning to provide reliable, automated tools for early colon cancer detection.