Colon Cancer Detection Using Deep Learning

Abstract

Colorectal cancer ranks as the third most dangerous cancer globally due to its rapid progression and significant impact on human body functions. Early detection is crucial for successful treatment outcomes before metastasis occurs. This study presents a comprehensive deep learning approach for automated colon cancer detection using histopathological images, emphasizing the critical role of image preprocessing in enhancing model performance. We implemented a modified VGG16 convolutional neural network combined with advanced image preprocessing methods, specifically bilateral filtering, to improve feature extraction and classification accuracy. The bilateral filter was selected for its unique ability to reduce noise while preserving important edge details in medical images, essential for accurate cancer detection. Our preprocessing pipeline includes bilateral filtering with optimized parameters (?s=?r=50), grayscale conversion with bone colormap enhancement, image resizing to 224 × 224 pixels, and normalization to [ 0, 1] range. The model was trained on the LC25000 dataset containing 1 0, 0 0 0 histopathological images equally distributed between benign tissue and adenocarcinoma classes. To prevent overfitting, we applied data augmentation including rotation, horizontal and vertical shifting, and random horizontal flipping. The training process utilized Adam optimizer with binary cross-entropy loss function, along with EarlyStopping and ReduceLROnPlateau callbacks. Our approach achieved remarkable results with 99.7 % accuracy, 99.8 % precision for non-cancerous tissue, and 100 % precision and recall for cancerous tissue detection. The study demonstrates that proper image preprocessing, particularly bilateral filtering, significantly enhances deep learning models for medical image analysis. These results suggest implementing this approach in computer-aided diagnosis systems for real-time colorectal cancer screening. © 2025 IEEE.