Parkinson's disease classification with CWNN: Using wavelet transformations and IMU data fusion for improved accuracy

Area of Science:

• Neuroscience and Biomedical Engineering
• Machine Learning for Healthcare

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder requiring early and accurate classification for effective treatment.
• Inertial Measurement Units (IMUs) offer a promising method for collecting movement data to aid in PD diagnosis.

Purpose of the Study:

• To develop and evaluate a Convolutional Wavelet Neural Network (CWNN) for classifying Parkinson's disease using IMU data.
• To identify the optimal combination of wavelet transform and IMU data type for maximizing PD classification accuracy.

Main Methods:

• Proposed a CWNN architecture integrating convolutional and wavelet neural networks to analyze spatial-temporal patterns in IMU data.
• Utilized Continuous Wavelet Transform (CWT) with various wavelet functions (Morlet, Mexican Hat, Gaussian).
• Trained and evaluated the CWNN using accelerometer, gyroscope, and fused IMU data.

Main Results:

• The CWNN model demonstrated robust performance in classifying PD patients.
• The combination of the Morlet wavelet function and fused IMU data achieved the highest classification accuracy.
• Performance was assessed using accuracy, precision, recall, and F1-score, highlighting the influence of wavelet choice and data type.

Conclusions:

• Combining CWT feature extraction with IMU data fusion in CWNNs significantly improves PD classification.
• Enhanced representation of PD-related movement patterns through CWT and data fusion leads to better diagnostic accuracy.
• This approach offers a promising avenue for developing more reliable and accurate PD diagnostic models.