DL-Based Automatic Detection and Classification for Endoscopy

Colorectal polyps are important precursors to colon cancer, which is the third most common cause of cancer mortality worldwide. During colonoscopy, physicians manually inspect the colon using a camera to search for polyps. Early detection is vital but challenging from an image processing standpoint. This study demonstrates that Convolutional Neural Networks (CNNs) can achieve over 90% accuracy in polyp detection within 2 minutes using small datasets with Keras data augmentation. Transfer learning with pre-trained VGG16 showed no improvement, but fine-tuning achieved approximately 98% accuracy.