Scholarlink

<p>Parkinson’s disease is a progressive neurodegenerative disorder that presents motor and non-motor impairments. Handwriting analysis offers a non-invasive and cost-effective alternative to conventional clinical diagnosis. Although deep learning methods have shown promising results for handwriting-based Parkinson’s disease detection, their performance is often affected by class imbalance and elevated false-negative rates, which are particularly harmful in medical diagnosis due to delayed intervention. To address this limitation, we propose a False Negative Driven Dynamic Weighted Cross-Entropy (FN-DDWCE) loss that adaptively penalises classes with higher false-negative occurrences. The proposed loss is integrated into a feature-level fusion model combining ResNet50 and MobileNetV2. Experiments on a six-class handwriting dataset achieve an accuracy of 95.01% with a reduced false-negative rate of 5.71%.</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

Dynamic false-negative-driven loss optimisation for the diagnosis of Parkinson’s disease using handwritten data

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Munzir Hubiba Bashir,
  • Tawseef Ayoub Shaikh

摘要

Parkinson’s disease is a progressive neurodegenerative disorder that presents motor and non-motor impairments. Handwriting analysis offers a non-invasive and cost-effective alternative to conventional clinical diagnosis. Although deep learning methods have shown promising results for handwriting-based Parkinson’s disease detection, their performance is often affected by class imbalance and elevated false-negative rates, which are particularly harmful in medical diagnosis due to delayed intervention. To address this limitation, we propose a False Negative Driven Dynamic Weighted Cross-Entropy (FN-DDWCE) loss that adaptively penalises classes with higher false-negative occurrences. The proposed loss is integrated into a feature-level fusion model combining ResNet50 and MobileNetV2. Experiments on a six-class handwriting dataset achieve an accuracy of 95.01% with a reduced false-negative rate of 5.71%.