Optimized Deep Learning for the Classification of Parkinson's Disease Based on Voice Features

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder.
• Early diagnosis of PD is crucial for effective management and treatment.
• Vocal features offer a non-invasive method for PD detection.

Purpose of the Study:

• To develop and validate a novel technique for diagnosing Parkinson's disease using voice signals.
• To enhance the accuracy and efficiency of PD detection through advanced signal processing and machine learning.

Main Methods:

• Voice signals were pre-processed to remove noise.
• Feature extraction included standard metrics and a novel Exponential delta-Amplitude Modulation Spectrogram (Exponential-delta AMS).
• Feature selection was performed using a hybrid Squirrel Search Water Algorithm (SSWA), and classification was achieved with an attention-based Long Short-Term Memory (LSTM) network trained by SSWA.

Main Results:

• The SSWA-based attention-based LSTM model achieved high diagnostic performance.
• The proposed method demonstrated 92.5% accuracy, 95.4% sensitivity, and 91.4% specificity.
• The novel Exponential-delta AMS feature and SSWA optimization contributed to improved results.

Conclusions:

• The developed voice analysis technique shows significant promise for early and accurate Parkinson's disease diagnosis.
• The integration of novel signal processing features and optimized machine learning models can enhance diagnostic capabilities.
• This approach offers a potential non-invasive tool for widespread PD screening.