摘要
精准的医学影像分割对临床诊断具有重要价值。然而,病灶在医学影像中常呈现形状不规则、边缘模糊等特征,给快速筛查与精准诊断带来了极大挑战。为此,该研究提出基于多尺度减法机制的网络模型M^(2)SNet,并将其应用于医学影像中的息肉检测。在五个息肉数据集上,与U-Net、Attention-UNet、TransUNet和ResUNet等主流模型进行对比实验,结果显示,M^(2)SNet展现出更优的分割性能。M^(2)SNet的核心创新在于,通过多尺度减法单元有效滤除特征冗余,强化不同层级特征间的互补性,进而实现病灶的精准定位与边界清晰分割。相较于传统U-Net,该模型在保持结构简洁性的同时,显著提升了边缘细节的刻画能力。
Accurate medical image segmentation is of great value for clinical diagnosis.However,lesions often present irregular shapes,blurred edges and other features in medical images,posing great challenges to rapid screening and accurate diagnosis.To this end,this study proposes a network model M^(2)SNet based on the multi-scale subtraction mechanism and applies it to polyp detection in medical images.On five polyp datasets,it conducts comparative experiments with mainstream models such as U-Net,Attention-UNet,TransUNet,ResUNet,and so on.The results show that M^(2)SNet shows better segmentation performance.The core innovation of M^(2)SNet is to effectively filter out feature redundancy through multi-scale subtraction units,strengthen the complementarity between different levels of features,and then achieve accurate positioning and clear boundary segmentation of lesions.Compared with the traditional U-Net,this model significantly improves the ability to depict edge details while maintaining structural simplicity.
作者
刘磊
周梦宇
LIU Lei;ZHOU Mengyu(Chinese People's Armed Police Force Sichuan Provincial Corps Hospital,Leshan 614000,China;Sichuan Province Zigong Fushun County Maternal and Child Health Hospital,Zigong 643200,China)
出处
《现代信息科技》
2025年第3期37-43,共7页
Modern Information Technology
