Automatic detection of Parkinsonian speech using wavelet scattering features.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neurology

Background:

• Parkinson's disease (PD) diagnosis can be challenging and often relies on subjective clinical assessments.
• Objective, non-invasive methods for early PD detection are crucial for timely intervention and management.
• Speech analysis offers a promising avenue for identifying subtle physiological changes associated with PD.

Purpose of the Study:

• To develop and evaluate an automated system for Parkinson's disease detection using speech signals.
• To investigate the efficacy of wavelet scattering networks and Fisher vectors for extracting relevant speech features for PD identification.
• To compare the performance of the proposed method against existing state-of-the-art techniques in PD detection.

Main Methods:

• Speech data from the PC-GITA database were analyzed.
• A two-layer wavelet scattering network was employed to generate locally stable and translation-invariant speech features.
• Fisher vectors were used to encode scattering features into fixed-size utterance-level vectors.
• Support vector machine (SVM) and feed-forward neural network (FFNN) classifiers were trained for binary classification (healthy vs. PD).

Main Results:

• The proposed method, utilizing wavelet scattering features and Fisher vector encoding, demonstrated superior performance compared to current state-of-the-art approaches.
• The best classification accuracy achieved was 87% for distinguishing individuals with Parkinson's disease from healthy controls.
• The system was particularly effective when analyzing speech from a text reading task.

Conclusions:

• Wavelet scattering networks combined with Fisher vectors provide a robust and effective feature extraction strategy for Parkinson's disease detection from speech.
• The developed automated system shows significant potential for early and accurate diagnosis of PD.
• This approach offers a promising, objective, and non-invasive tool to aid clinicians in Parkinson's disease diagnosis.