Histopathological cancer detection based on deep learning and stain images

Abstract

Colorectal cancer (CRC)-a malignant growth in the colon or rectum- is the second largest cause of cancer deaths worldwide. Early detection may increase therapy choices. Deep learning can improve early medical detection to reduce the risk of unintentional death from an incorrect clinical diagnosis. Histopathological examination of colon cancer is essential in medical research. This paper proposes a deep learning-based colon cancer detection method using stain-normalized images. We use deep learning methods to improve detection accuracy and efficiency. Our solution normalizes image stain variations and uses deep learning models for reliable classification. This research improves colon cancer histopathology analysis, which may enhance diagnosis. Our paper uses DenseNet-121, VGG-16, GoogLeNet, ResNet-50, and ResNet-18 deep learning models. We also analyze how stain normalization (SN) improves our model on histopathology images. The ResNet-50 model with SN yields the highest values (9.94%) compared to the other four models and the nine models from previous studies.