MPED-YOLO:多尺度息肉影像检测算法OA

Accurate polyp detection is crucial for the diagnosis and treatment of colorectal cancer.However,existing methods are prone to detail loss during downsampling and scale mismatch in feature extraction,making it difficult to identify key mucosal features of polyps.To overcome these limitations,we propose a polyp detail-enhanced detection algorithm based on MPED-YOLO(Multi-scale Enhanced Detection-YOLO).The proposed Sparse Enhanced Residual Module(SERM)expands the receptive field to improve multi-scale polyp feature extraction.A Mucosal Detail Retention Module(MDRM)optimizes spatial and depth features,focusing on lesion fine-grained a-nalysis while reducing model sensitivity to interferences(e.g.,mucus adhesion,intestinal fold artifacts).Additionally,a Boundary Multi-Scale Feature Fusion Module(BMFFM)dynamically adjusts feature map weights for adaptive multi-scale fusion,further enhan-cing small polyp detection accuracy.Experimental results on two public datasets,CVC-ClinicDB and Kvasir-SEG,show that MPED-YOLO achieves mAP@0.5 of 85.9%and 95.2%,respectively,which are 11.3%and 4.3%higher than that of the baseline model.In addition,the proposed model achieves mAP@0.5 of 97.3%on a private clinical dataset.MPED-YOLO exhibits excellent accuracy and robustness in polyp detection,verifying its potential value in practical clinical applications and providing a new approach to improve the accuracy and efficiency of gastrointestinal polyp diagnosis.