Scholarlink

<p>Osteoporosis is underdiagnosed because dual-energy X-ray absorptiometry (DXA) is costly and scarce. We present <b>MultiScaleKANNet</b>, a hybrid deep-learning architecture for radiographic bone-loss risk stratification from routine knee X-rays, combining convolutional feature learning, learnable nonlinear transformations via Kolmogorov–Arnold Network (KAN) layers, and Transformer-based multi-scale attention. We evaluated MultiScaleKANNet on 2,035 knee radiographs from four public Kaggle sources with three categories (Healthy, Osteopenia, Osteoporosis). The labels are <i>proxy labels</i>—some derived from quantitative ultrasound T-scores rather than DXA—so results represent radiographic risk stratification, not clinical diagnosis. On a stratified held-out test set (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(n{=}407\)</EquationSource> </InlineEquation>), the model achieved 97.30% accuracy (95% CI: 95.3–98.6%; Cohen’s <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\kappa {=}0.9584\)</EquationSource> </InlineEquation>; MCC<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\({=}0.9585\)</EquationSource> </InlineEquation>; micro-averaged AUC<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\({=}0.9749\)</EquationSource> </InlineEquation>). A source-held-out evaluation yielded 89.52% binary accuracy (<InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\kappa {=}0.7903\)</EquationSource> </InlineEquation>), suggesting in-distribution metrics may partly reflect dataset homogeneity. Ablation studies confirm synergistic gains from KAN layers (+2.46%), multi-scale processing (+4.17%), and Transformer attention (+4.91%), with 40% parameter reduction versus ResNet-18. This is a <i>methodological feasibility study</i>; prospective DXA-confirmed validation is required.</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

MultiScaleKANNet: a hybrid CNN-KAN-transformer architecture for radiographic bone-loss risk stratification from knee X-rays

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月12日

查看全文
  • Ahmed S. Shaban,
  • Mohammed Tawfik,
  • Islam Fathi,
  • Ayman Myla

摘要

Osteoporosis is underdiagnosed because dual-energy X-ray absorptiometry (DXA) is costly and scarce. We present MultiScaleKANNet, a hybrid deep-learning architecture for radiographic bone-loss risk stratification from routine knee X-rays, combining convolutional feature learning, learnable nonlinear transformations via Kolmogorov–Arnold Network (KAN) layers, and Transformer-based multi-scale attention. We evaluated MultiScaleKANNet on 2,035 knee radiographs from four public Kaggle sources with three categories (Healthy, Osteopenia, Osteoporosis). The labels are proxy labels—some derived from quantitative ultrasound T-scores rather than DXA—so results represent radiographic risk stratification, not clinical diagnosis. On a stratified held-out test set ( \(n{=}407\) ), the model achieved 97.30% accuracy (95% CI: 95.3–98.6%; Cohen’s \(\kappa {=}0.9584\) ; MCC \({=}0.9585\) ; micro-averaged AUC \({=}0.9749\) ). A source-held-out evaluation yielded 89.52% binary accuracy ( \(\kappa {=}0.7903\) ), suggesting in-distribution metrics may partly reflect dataset homogeneity. Ablation studies confirm synergistic gains from KAN layers (+2.46%), multi-scale processing (+4.17%), and Transformer attention (+4.91%), with 40% parameter reduction versus ResNet-18. This is a methodological feasibility study; prospective DXA-confirmed validation is required.