Early Detection of Colorectal Cancer from Polyps Images Using Deep Learning

Detecting colorectal cancer in early stages and predicting associated risk is very important as it has the highest mortality rate after lung cancer. The polyps present in the colon and rectum are one of the main causes of colorectal cancer. In this study, we propose different deep learning methods to classify images as normal or containing polyps. The performance of these models was evaluated via implementation on public and private datasets. On the WCE dataset, the OEM model implemented with transfer learning achieved a classification accuracy of 99.5%, 96.37%, and 96% on training, validation, and test set, respectively. On the Kvasir dataset, the OEM model obtained a classification accuracy of 97.80%, 96.40%, and 96.03% on training, validation, and test set, respectively. On SCPolyps dataset, the OEM model achieved a classification accuracy of 93.80%. Hence, the utilization of deep learning techniques can be viewed as an efficient approach to creating tools for assisting in the diagnosis of CRC.