| Abstrakt: |
Colon Cancer (CoC) appears to be the third leading cause of cancer death in men and second among women. Therefore, the infection and mortality rates can be reduced with early identification of this cancer. For accurate colon cancer screening, this research proposes a novel deep-learning strategy using Residual Networks (ResNet). In this paper, the performance of four architectures, CoC-ResNet152, CoC-ResNet101, CoC-ResNet50, and CoC-ResNet50V2, is evaluated. Different deep learning architectures are compared in detail using the two-class public LC25000 histopathologic datasets, namely benign and Adenocarcinoma. From the experimental data, it is determined that the CoC-ResNet50V2 model is superior to the other three proposed models. After training on histopathological images, the CoC-ResNet50V2 model achieved 99.55% accuracy, 99.41% specificity, 99.38% precision, 99.69% recall, 0.58% False Positive Rate (FPvR), 0.30% False Negative Rate (FNvR), 99.54% F1 score, 0.990 Matthew's Correlation Coefficient (MCC), and 0.992 Cohen's Kappa Coefficient (KP). [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink