Scholarlink

<p>Preoperative differentiation of Acute Appendicitis (AA), Chronic Appendicitis (CA), and the clinically challenging Acute Exacerbation of Chronic Appendicitis (AEC) remains a significant diagnostic dilemma. This study aimed to develop a multi-modal deep learning framework integrating ultrasound (US) images with domain-specific features to enhance diagnostic precision. In this retrospective study, we analyzed 605 pathology-confirmed patients (392 AA, 150 CA, 63 AEC). The dataset comprised preoperative US images and a complete 19-dimensional feature vector encompassing clinical metrics and handcrafted sonographic markers guided by clinical protocols. We developed a multi-modal Two-Stream fusion framework, termed TSNet, utilizing a ResNet-50 backbone with Spatial Attention (SPA) modules for visual extraction, fused with tabular data. Class imbalance was addressed using Focal Loss, and performance was evaluated via 5-fold stratified cross-validation. Results indicated a performance trade-off between global precision and subtype-specific sensitivity. While the single-modal ResNet-SPA achieved the highest Overall Accuracy (<i>0.8612</i>) and Macro PR-AUC (<i>0.7369</i>), the multi-modal <i>TSNet demonstrated superior discriminative robustness</i> across all subtypes. Specifically, <i>TSNet attained the highest Macro-average ROC-AUC of 0.8901</i>, significantly enhancing the detection of the clinically elusive AEC subtype. By prioritizing sensitivity, TSNet achieved an <i>AEC Recall of 47.6%</i> (vs. 41.3% for ResNet-SPA) and the highest AEC F1-score of <i>0.4800</i>. While this improvement in AEC sensitivity is modest in absolute terms, it is clinically promissing.In conclusion, TSNet effectively differentiates appendicitis subtypes by leveraging the complementary strengths of deep visual representations and expert clinical knowledge. However, these findings require confirmation in larger, multi-center studies before influencing routine decision-making.</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

TSNet: a multi-modal deep learning framework for subtyping appendicitis: integrating ultrasound images, handcrafted features, and clinical data

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Zhuanghe He,
  • Rong Ma,
  • Xinyi Huang,
  • Liren Yang,
  • Erqing Liao,
  • Yang Chen

摘要

Preoperative differentiation of Acute Appendicitis (AA), Chronic Appendicitis (CA), and the clinically challenging Acute Exacerbation of Chronic Appendicitis (AEC) remains a significant diagnostic dilemma. This study aimed to develop a multi-modal deep learning framework integrating ultrasound (US) images with domain-specific features to enhance diagnostic precision. In this retrospective study, we analyzed 605 pathology-confirmed patients (392 AA, 150 CA, 63 AEC). The dataset comprised preoperative US images and a complete 19-dimensional feature vector encompassing clinical metrics and handcrafted sonographic markers guided by clinical protocols. We developed a multi-modal Two-Stream fusion framework, termed TSNet, utilizing a ResNet-50 backbone with Spatial Attention (SPA) modules for visual extraction, fused with tabular data. Class imbalance was addressed using Focal Loss, and performance was evaluated via 5-fold stratified cross-validation. Results indicated a performance trade-off between global precision and subtype-specific sensitivity. While the single-modal ResNet-SPA achieved the highest Overall Accuracy (0.8612) and Macro PR-AUC (0.7369), the multi-modal TSNet demonstrated superior discriminative robustness across all subtypes. Specifically, TSNet attained the highest Macro-average ROC-AUC of 0.8901, significantly enhancing the detection of the clinically elusive AEC subtype. By prioritizing sensitivity, TSNet achieved an AEC Recall of 47.6% (vs. 41.3% for ResNet-SPA) and the highest AEC F1-score of 0.4800. While this improvement in AEC sensitivity is modest in absolute terms, it is clinically promissing.In conclusion, TSNet effectively differentiates appendicitis subtypes by leveraging the complementary strengths of deep visual representations and expert clinical knowledge. However, these findings require confirmation in larger, multi-center studies before influencing routine decision-making.