Parkinson's Disease Action Tremor Detection with Supervised-Leaning Models.

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Parkinson's Disease (PD) significantly impacts patients' quality of life due to symptoms like hand tremors.
• Accurate tremor detection is crucial for developing effective wearable devices to manage PD symptoms.
• Existing detection methods require improvement for real-time application.

Purpose of the Study:

• To introduce a novel method for detecting action tremors in Parkinson's Disease (PD).
• To differentiate PD tremors from normal daily activities using sensor data.
• To evaluate the performance of machine learning models for PD tremor detection.

Main Methods:

• Utilized accelerometer and gyroscope data from 30 PD patients wearing wrist sensors.
• Extracted time-domain and frequency-domain hand-crafted features.
• Compared hand-crafted features with Convolutional Neural Network (CNN) features.
• Trained and evaluated Logistic Regression (LR), K-Nearest Neighbours (KNNs), Support Vector Machines (SVMs), and CNNs.

Main Results:

• Hand-crafted features showed clearer distinctions in t-SNE visualization compared to CNN features.
• All evaluated models achieved over 90% F1 scores in five-fold cross-validation.
• Support Vector Machines (SVMs) demonstrated superior performance with over 92% F1 score in cross-validation and over 90% in leave-one-out evaluation.

Conclusions:

• The proposed feature extraction method effectively detects Parkinson's Disease action tremors.
• Machine learning models, particularly SVMs, show high efficacy in identifying PD tremors.
• This research contributes to the advancement of wearable technology for Parkinson's Disease management.