EfficientNetB2-Based Deep Learning Model for Differentiating Colorectal Adenomatous and Hyperplastic Tissue Images

Colorectal cancer (CRC) is one of the most common causes of cancer death worldwide, which makes prompt and precise identification of precancerous polyps such as adenomatous and hyperplastic types essential for effective treatment. This study examines how well different EfficientNet architectures (B0 to B4) classify colorectal polyps, using the publicly accessible Colon Cancer dataset. In order to guarantee an objective comparison, all models were trained under the same conditions, using the same preprocessing and augmentation strategies. The evaluation metrics comprised F1 score, precision, and recall. The EfficientNetB2 model outperformed the others in distinguishing between adenomatous and hyperplastic polyps. The model produced results for hyperplastic precision value of 0.99, recall value of 0.98, and F1 score of 0.99, as well as adenomatous precision value is 0.98, recall value of 0.99, and F1-score of 0.99. These results show that EfficientNet-B2 provides the best balance between model depth and classification accuracy, positioning it as a promising tool for automated CRC screening. Future studies will aim to combine classification with real-time segmentation models to aid clinical decision-making.