Scholarlink

<p>Diabetic foot ulcers (DFU) are a severe complication often leading to non-traumatic amputations if caught too late. While infrared thermography acts as a tool for early intervention by detecting hidden thermal anomalies, traditional AI models frequently act as black boxes that treat patients in isolation. This study introduces a shift toward Relational Manifold Learning through the <i>k</i>-GCNDFU framework. By transforming patient profiles into a connected Patient Similarity Network (PSN), our approach captures deep structural dependencies across a balanced manifold of 753 unique instances. Our six-layer Graph Convolutional Network (GCN) achieved a classification accuracy of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(89.7\% \pm 2.7\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>89.7</mn> <mo>%</mo> <mo>±</mo> <mn>2.7</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>, a mean AUC of <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(0.952 \pm 0.025\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>0.952</mn> <mo>±</mo> <mn>0.025</mn> </mrow> </math></EquationSource> </InlineEquation>, and a Matthews Correlation Coefficient (MCC) of <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(0.782 \pm 0.059\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>0.782</mn> <mo>±</mo> <mn>0.059</mn> </mrow> </math></EquationSource> </InlineEquation>. Furthermore, GNNExplainer provides transparency by identifying the Medial Calcaneal Artery (MCA) as a key physiological driver of detection. These results demonstrate that the proposed relational approach offers a consistent, reliable and interpretable tool for the early detection of diabetic foot complications.</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

k–GCNDFU: An explainable relational manifold learning framework for diabetic foot ulcer classification

  • 期刊论文

  • 发表日期：

    2026年4月25日

查看全文
  • Parul Chauhan,
  • Geeta Sikka,
  • Chandra Prakash

摘要

Diabetic foot ulcers (DFU) are a severe complication often leading to non-traumatic amputations if caught too late. While infrared thermography acts as a tool for early intervention by detecting hidden thermal anomalies, traditional AI models frequently act as black boxes that treat patients in isolation. This study introduces a shift toward Relational Manifold Learning through the k-GCNDFU framework. By transforming patient profiles into a connected Patient Similarity Network (PSN), our approach captures deep structural dependencies across a balanced manifold of 753 unique instances. Our six-layer Graph Convolutional Network (GCN) achieved a classification accuracy of \(89.7\% \pm 2.7\%\) 89.7 % ± 2.7 % , a mean AUC of \(0.952 \pm 0.025\) 0.952 ± 0.025 , and a Matthews Correlation Coefficient (MCC) of \(0.782 \pm 0.059\) 0.782 ± 0.059 . Furthermore, GNNExplainer provides transparency by identifying the Medial Calcaneal Artery (MCA) as a key physiological driver of detection. These results demonstrate that the proposed relational approach offers a consistent, reliable and interpretable tool for the early detection of diabetic foot complications.