Deep learning based analysis in oncological studies: Colorectal cancer staging

Abstract

The introduction of deep learning in the medical sector for the diagnosis and classification of tissue images has proven to have the best accuracy level over the years. Hence in this study, the utilization of deep learning in colorectal cancer staging as well as the use of AlexNet Architecture in the training of datasets were used in the prediction and detection of colorectal cancer. Moreover, the novelty of the technique used in this study involved a direct analysis of the dataset. Colorectal cancer was analyzed using deep learning technique and AlexNet architecture (CNN) from a dataset of 7180 from 50 different patients having colorectal cancer, which eventually produced 900 images. The results revealed that deep learning was analyzed on several classes: adipose tissue, back tissue, debris tissue, lymphocytes tissues, mucus tissue, smooth muscle, normal tissue, cancer associated stroma and adenocarcinoma epithelium, which had an accuracy of 97.8% for all the classes, (1, 1, 0.952, 0.952, 1, 1, 0.952, and 0.952) respectively, other classes had a precision of 1 except adenocarcinoma epithelium with 0.95, (1, 1, 0.975, 0.975, 1, 1, 0.975, and 0.95) respectively for the F1-score. Hence the result revealed that deep-learning technique for colorectal cancer is very effect.