سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

Home / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

HEALTH: Hyperbolic Embedding and Acoustic-based Learning for Topological Hierarchies in Parkinson’s Disease

Authors :

Saghar Shafaati¹ S. Hossein Erfani²

1- Department of Computer Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran. 2- Department of Computer Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran.

Keywords :

Parkinson’s disease،Hyperbolic embedding،Acoustic biomarkers،Explainable AI،Disease progression modeling

Abstract :

Parkinson's disease (PD) is a neurodegenerative disorder characterized by heterogeneous motor and non-motor features complicating early diagnosis and individualized monitoring. Recent reports have identified acoustic biomarkers to be non-invasive prodromal PD predictors, but classical modeling approaches often fail to capture the complex, hierarchical nature of disease progression. This study introduces HEALTH (Hyperbolic Embedding and Acoustic-based Learning for Topological Hierarchies), a novel computational framework that integrates graph-based similarity modeling, hyperbolic geometry, unsupervised clustering, and explainable supervised classification to characterize dysarthric speech patterns in PD. Sustained phonation recordings from participants were preprocessed and embedded in a two-dimensional Poincaré disk, wherein hyperbolic distances reflected latent acoustic dissimilarities. The embedding optimization achieved a ~95% reduction in reconstruction loss, with silhouette coefficients stabilizing near 0.44, indicating robust cluster separation. SHAP analysis identified pitch entropy, amplitude variability, and frequency-related measures as principal determinants of classification outcomes, supporting the clinical interpretability of the model. Comparative evaluation demonstrated that HEALTH outperforms traditional Euclidean approaches in both stratification and explainability. This work underscores the potential of hyperbolic embeddings as scalable, interpretable tools for precision monitoring of neurodegenerative disease and contributes a reproducible methodology to advance non-invasive, data-driven diagnostics in PD.