A Robust Colon Cancer Detection Model Using Deep- Learning

Colon cancer is the third most common cancer, with a high mortality rate and a high pathological need for early and accurate diagnosis. Histopathology is a golden standard tool for diagnosing almost all types of cancer, including colon cancer. It can be enhanced using deep learning techniques if the key challenges, namely, clean data annotation and inter-observer variability, are addressed. This highlights a need for a noise-robust algorithm with computational feasibility to work on real-world clinical data. In this paper, the experiment uses different deep learning models- ResNet-34, XCiT, SqueezeNet, MobileNet-individually integrated with a data cleaning architecture in an attempt to (1) achieve unmatched accuracy, (2) draw a comparison amongst the performance of these models and (3) point to the best model as a solution to the given problem. MobileNet outperforms all other models as the backbone for the algorithm, achieving the highest accuracy (84.39%), thus boosting the algorithm performance on the Chaovang dataset.