Cancer colon is still considered one of the major health problems in the world today because of the high level of incidence and mortality and therefore calls for early and accurate diagnosis. Conventional screening processes are insufficient and imprecise for early detection consequently leading to the use of better mathematical modeling and computational tools. This study proposes a deep learning framework leveraging two specialized datasets: KVASIR dataset, which comprises of a range of gastrointestinal endoscopy image samples; and the ETIS-Larib-Polyp DB dataset with special emphasis on colorectal polyps. To increase diagnostic accuracy, we used the DenseNet201, VGG19, and two Custom CNNs: CNN01 and CNN02. Additional image preprocessing measures including resizing the image to a manageable size, normalizing the image, and filtering of the images also enhanced the performance of the models. The results show that DenseNet201 and CNN02 outperformed the other models, with accuracy scores of 99.22% and 99.37%, respectively. This study demonstrates how deep learning has the potential to revolutionize colon cancer detection by allowing more precise and early diagnosis, thus reducing healthcare costs and improving the health of patients