Nested U-Net-Based Polyp Segmentation for Accurate Diagnosis of Colorectal Cancer

Colorectal cancer, a prominent global malignancy characterized by abnormal polyp growth in the colon or rectum, presents significant challenges in accurate diagnosis and treatment. Manual polyp segmentation during colonoscopy is hampered by obscured visibility, varying polyp morphology, and potential human error. To address these limitations, automated segmentation methodologies have emerged as crucial tools for improving diagnostic precision and efficiency. While conventional machine learning techniques have been applied to alleviate these complexities, the emergence of deep learning has opened new avenues for intricate pattern recognition and feature extraction. This research proposes an innovative approach to polyp segmentation, harnessing the Nested U-Net architecture complemented by a combination loss function. The proposed methodology strategically integrates augmentation techniques to counteract overfitting, a common concern in deep learning applications. Evaluation metrics such as dice coefficient, Intersection over Union (IoU), precision, and recall are employed to measure the performance of the proposed approach. Compared to U-Net-based polyp segmentation, the proposed approach exhibits a substantial increase of 0.60% in training dice coefficient, 1.15% in training IoU and 0.01% in precision. The results of extensive experimentation demonstrate the superiority of the proposed Nested U-Net-based polyp segmentation over U-Net-based approaches.