MMDD-Ensemble: A Multimodal Data-Driven Ensemble Approach for Parkinson's Disease Detection.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Parkinson's disease (PD) is the second most prevalent neurological disorder.
• Current diagnostic methods lack a specific medical test, necessitating advanced detection techniques.

Purpose of the Study:

• To identify optimal voice data modalities and features for enhanced PD detection.
• To propose and validate a MultiModal Data-Driven Ensemble (MMDD-Ensemble) approach for PD detection using voice data.

Main Methods:

• Collected multimodal voice data via Smart Phone (SP) and Acoustic Cardioid (AC) channels.
• Utilized sustained phonation (P), speech (S), voiced (V), and unvoiced (U) modalities.
• Developed a two-level MMDD-Ensemble: base classifiers followed by blending and voting fusion.

Main Results:

• The MMDD-Ensemble achieved 96% accuracy, 100% sensitivity, 88.88% specificity, 0.914 MCC, and 0.986 AUC.
• Outperformed optimal unimodal frameworks and state-of-the-art ensemble models in accuracy and AUC.
• Demonstrated the robustness and effectiveness of the multimodal approach for PD detection.

Conclusions:

• The proposed MMDD-Ensemble method shows significant promise for accurate and reliable Parkinson's disease detection.
• Multimodal voice analysis offers a powerful, non-invasive strategy for neurological disorder diagnosis.
• Further research can refine this approach for clinical application.