Scholarlink

Purpose <p>To develop and validate a robust, interpretable biocybernetic system for quantitative Parkinson’s disease (PD) assessment from voice recordings, translating vocal disruptions into a continuous physiological metric.</p> Methods <p>We propose a dual-stream architecture fusing deep contextual representations with 40 peripheral neuromotor biomarkers via Multi-Modal Cross-Attention Fusion (MM-CAF). The system was trained on Spanish speech using uncertainty-aware multi-task learning with learnable loss weighting, and externally validated on Italian speech without retraining. A strict subject-independent stratified 5-fold cross-validation was employed.</p> Results <p>On the Spanish cohort (<i>n</i> = 126), the system achieved accuracy 90.5% ± 7.3%, F1-score 0.898, and AUC 0.95 ± 0.04 (95% CI: 0.91–0.99). External validation on Italian (<i>n</i> = 50) confirmed cross-linguistic generalizability (AUC: 0.93; accuracy: 78.0%). Severity tracking on the independent UCI cohort (<i>n</i> = 5,875) showed significant correlation with UPDRS-III (Pearson <i>r</i> = 0.564, <i>p</i> &lt; 0.001), with MAE of 5.58 points.</p> Conclusion <p>This biocybernetic system captures language-invariant pathophysiological features and provides a continuous, uncertainty-calibrated severity score for longitudinal monitoring, representing an advance in voice-based physiological measurement for remote PD management.</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

A Biocybernetic System for Quantifying Parkinsonian Motor Impairment from Voice with Uncertainty-Aware Closed-Loop Monitoring

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Leiyong Guo,
  • Bo Zhou,
  • Degang Xing

摘要

Purpose

To develop and validate a robust, interpretable biocybernetic system for quantitative Parkinson’s disease (PD) assessment from voice recordings, translating vocal disruptions into a continuous physiological metric.

Methods

We propose a dual-stream architecture fusing deep contextual representations with 40 peripheral neuromotor biomarkers via Multi-Modal Cross-Attention Fusion (MM-CAF). The system was trained on Spanish speech using uncertainty-aware multi-task learning with learnable loss weighting, and externally validated on Italian speech without retraining. A strict subject-independent stratified 5-fold cross-validation was employed.

Results

On the Spanish cohort (n = 126), the system achieved accuracy 90.5% ± 7.3%, F1-score 0.898, and AUC 0.95 ± 0.04 (95% CI: 0.91–0.99). External validation on Italian (n = 50) confirmed cross-linguistic generalizability (AUC: 0.93; accuracy: 78.0%). Severity tracking on the independent UCI cohort (n = 5,875) showed significant correlation with UPDRS-III (Pearson r = 0.564, p < 0.001), with MAE of 5.58 points.

Conclusion

This biocybernetic system captures language-invariant pathophysiological features and provides a continuous, uncertainty-calibrated severity score for longitudinal monitoring, representing an advance in voice-based physiological measurement for remote PD management.