Shweta Gangrade

and 4 more

Colorectal cancer is responsible for a high proportion of cancer mortality. The most effective way to avoid colorectal cancer is to have a colonoscopy. However, not every polyp in the colon is prone to cancer. As a result, different techniques are employed to classify polyps. A video endoscopy can diagnose stomach ulcers, bleeding, and polyps. Doctors spend a lot of time reviewing medical video endoscopy images. The challenge of diagnosing images by manually has spurred research into computer-assisted methods that can accurately and swiftly assess any created image. The suggested approach develops a framework for identifying digestive problems. The methods and treatment plan would be determined by the gastrointestinal state classification. In the present study, publicly accessible datasets, such as Kvasir, in used. In the Kvasir dataset, 5000 images are evenly distributed across five different digestive tract-related categories: ulcerative colitis, dye-lifted polyps, resection margins, normal cecum, and polyps. Preprocessing is done to improve the quality of the images and reduce the noise. These improved images were employed using deep learning networks. Present study proposes a stacking ensemble approach to boost the model's accuracy for prediction. The ensemble approach included five meticulously tuned deep convolutional neural network architectures, namely Xception, ResNet-101, VGG-19, EfficientNetB2v3, and MobineNetV2. These models were trained using weights obtained from the ImageNet dataset. Highest accuracy of 96.50% was achieved using meta models based on K-nearest neighbour (K-NN) method