摘要
皮肤病的发病率在全球呈上升趋势,它影响了患者的生活质量,甚至可能导致睡眠障碍、抑郁症等心理健康问题。目前,现有模型在皮肤病变图像语义分割方面表现不佳。因此构建了高质量的数据集CliAD,该数据具有精准的标注,并且由于其高度的真实性,图片中包含大量点状标注,可以看到不同类别的皮肤病之间存在显著差异,为病变的识别与分割带来了诸多挑战。为了应对这些挑战,提出了语义分割模型MVMNet。该模型基于Mamba和Convolutional Neural Network,采用VMM Blcoks进行细粒度特征提取,提高了对点状区域的识别能力。为了解决类别间的差异,使用U形结构在不同层次融合和提取特征,以识别每个类别的特征。实验结果表明,模型在CliAD、ISIC17和ISIC18数据集上表现优异。
The incidence of skin diseases is rising globally,impacting patients′quality of life and potentially leading to mental health issues such as sleep disorders and depression.Existing models exhibit suboptimal performance in semantic segmentation of dermatological lesion imag⁃es.This paper therefore constructs the high-quality CliAD dataset,featuring precise annotations and highly realistic images containing numerous punctate annotations.Significant differences between dermatological categories present substantial challenges for lesion identification and seg⁃mentation.To address these challenges,the semantic segmentation model MVMNet is proposed.This model is based on Mamba and Convolutional Neural Network(CNN),utilizing VMM Blocks for fine-grained feature extraction to enhance recognition of punctate regions.To address inter-cat⁃egory differences,a U-shaped architecture is employed to fuse and extract features across different levels,enabling the identification of category specific characteristics.Experimental results demonstrate the model′s superior performance on the CliAD,ISIC17,and ISIC18 datasets.
作者
彭健强
张宇凡
PENG Jianqiang;ZHANG Yufan(China Telecom Corporation Limited Sichuan Branch,Chengdu 610072,China;Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《通信与信息技术》
2025年第S1期1-4,共4页
Communication & Information Technology
