Scholarlink

<p>This paper presents Tongue-U-Net, a novel multimedia analysis framework for automated disease diagnosis through tongue image processing. The proposed system integrates dual attention mechanisms with U-Net architecture to address challenges in medical image analysis, particularly for tongue image segmentation and classification. Our multimedia tool processes tongue images to segment seven anatomical regions and classify four disease categories (digestive, cardiovascular, respiratory, renal) with high algorithmic performance under controlled experimental conditions. Experimental results on multi-ethnic datasets (300 Chinese and 951 Iranian images) demonstrate strong algorithmic performance, achieving 97.98% ± 0.6% Dice coefficient for segmentation and 97.5% ± 0.5% accuracy for classification. The framework’s robustness, validated through 5-fold cross-validation, highlights its potential as a reliable multimedia tool for automated disease screening and decision support. This research contributes significant advancements to multimedia applications in healthcare by providing an efficient, accurate, and scalable solution for automated disease screening.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Tongue-U-Net: A dual-attention multimedia framework for automated disease diagnosis via tongue image analysis

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Hamed Aghapanah,
  • Morteza Choubin,
  • Amir Asadi

摘要

This paper presents Tongue-U-Net, a novel multimedia analysis framework for automated disease diagnosis through tongue image processing. The proposed system integrates dual attention mechanisms with U-Net architecture to address challenges in medical image analysis, particularly for tongue image segmentation and classification. Our multimedia tool processes tongue images to segment seven anatomical regions and classify four disease categories (digestive, cardiovascular, respiratory, renal) with high algorithmic performance under controlled experimental conditions. Experimental results on multi-ethnic datasets (300 Chinese and 951 Iranian images) demonstrate strong algorithmic performance, achieving 97.98% ± 0.6% Dice coefficient for segmentation and 97.5% ± 0.5% accuracy for classification. The framework’s robustness, validated through 5-fold cross-validation, highlights its potential as a reliable multimedia tool for automated disease screening and decision support. This research contributes significant advancements to multimedia applications in healthcare by providing an efficient, accurate, and scalable solution for automated disease screening.