SS-DRPL: self-supervised deep representation pattern learning for voice-based Parkinson's disease detection.

Area of Science:

• Neurology
• Computer Science
• Biomedical Engineering

Background:

• Parkinson's disease (PD) poses a global health challenge, requiring accurate diagnostic tools.
• Voice analysis offers a non-invasive method for PD detection.
• Self-supervised deep representation pattern learning (SS-DRPL) shows promise for enhancing data representation.

Purpose of the Study:

• To investigate the effectiveness of SS-DRPL combined with deep learning for voice-based PD classification.
• To evaluate and compare the performance of different deep learning models (LSTM-RNN, DNN) augmented with SS-DRPL.
• To establish a performance baseline using traditional machine learning models.

Main Methods:

• Utilized SS-DRPL to extract features from voice data.
• Implemented and compared hybrid Long Short-Term Memory and Recurrent Neural Networks (LSTM-RNN) and Deep Neural Networks (DNN) architectures.
• Integrated SS-DRPL with deep learning models for PD classification.
• Included traditional machine learning models for comparative analysis.

Main Results:

• SS-DRPL integration improved model performance across all experimental setups.
• The LSTM-RNN architecture combined with SS-DRPL achieved the highest F1-score of 0.94.
• Deep learning models augmented with SS-DRPL demonstrated superior accuracy in voice-based PD detection.

Conclusions:

• SS-DRPL is an effective technique for enhancing deep learning models in voice-based PD detection.
• The LSTM-RNN architecture with SS-DRPL shows significant potential for accurate and efficient PD classification.
• This approach offers a promising avenue for improving early diagnosis and intervention strategies for Parkinson's disease.