Telemonitoring Parkinson's disease using machine learning by combining tremor and voice analysis

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Parkinson's disease (PD) diagnosis is challenging in developing countries due to limited resources and symptom variability.
• Timely and accurate PD detection is crucial for effective management and treatment.
• Existing diagnostic methods may not be accessible or suitable for remote monitoring in underserved regions.

Purpose of the Study:

• To develop and evaluate a cloud-based machine learning system for remote Parkinson's disease detection.
• To combine multiple PD indicators, specifically rest tremor and voice degradation, for improved diagnostic accuracy.
• To facilitate telemonitoring of PD patients in developing countries using accessible smartphone technology.

Main Methods:

• Utilized smartphone sensors (accelerometer and voice recorder) to collect rest tremor and vowel phonation data.
• Trained and optimized machine learning models using data from diagnosed PD patients and healthy individuals.
• Employed ensemble averaging on majority-vote predictions from k-nearest neighbors (kNN), support vector machine (SVM), and naive Bayes (NB) algorithms for final PD detection.

Main Results:

• Individual analysis of voice and tremor data achieved high accuracies of [Formula: see text] and [Formula: see text], respectively.
• k-nearest neighbors (kNN) demonstrated superior performance over SVM and NB for offline data analysis.
• Ensemble averaging of combined voice and tremor data analysis resulted in an average PD detection accuracy of [Formula: see text].

Conclusions:

• The proposed cloud-based system effectively detects Parkinson's disease using smartphone-captured voice and tremor data.
• This system enhances accessibility to medical diagnosis for aging populations in developing countries, particularly during situations limiting in-person consultations.
• The technology supports continuous patient monitoring and potential model updates through ongoing data collection.