COP-Detect: An End-to-End Transformer based Enhanced Colon Polyp Detection and Segmentation Framework Using Colonoscopy Images

Detecting polyps poses a significant challenge for the automated study of endoscopic images in computer-assisted clinical support systems.Our proposed deep learning approach, COP-Detect, addresses these issues through a series of advanced techniques. We start by preprocessing colonoscopy images using essential steps such as resizing, normalization, and data augmentation techniques, including geometric transformations and photometric adjustments.Dual-Group Feature Amplification Transformer (DG-Trans) this component enhances feature extraction by providing dual-adjust feature enrichment and group-wise feature enhancement, capturing both shallow and deep features critical for effective segmentation. Group-wise enhancement improves feature extraction within specific channels.In the decoder path, the Refined-Up Sampling Module (R-UP) restores detailed information in the segmentation maps, while the Adaptive Receptive-Field Detection Layer (ARD) uses atrous convolutions to capture contextual details surrounding polyps. Finally, the Deep Supervision Spatial Segmentation Layer (D3SL) includes auxiliary segmentation heads at various stages, predicting segmentation maps alongside the main output to enhance overall model performance.