Multi-label speech feature selection for Parkinson's Disease subtype recognition using graph model.

Area of Science:

• Neuroscience
• Speech Pathology
• Machine Learning

Background:

• Parkinson's Disease (PD) is a common neurodegenerative disorder with known links to dysphonia.
• Speech signal analysis is effective for PD identification and severity prediction.
• Recognizing PD subtypes (e.g., tremor, freezing of gait, dysphagia) is crucial for tailored diagnosis and treatment.

Purpose of the Study:

• To develop a multi-label learning framework for simultaneous recognition of Parkinson's Disease subtypes using speech signals.
• To investigate the efficacy of a novel graph-based multi-label feature selection algorithm for PD subtype identification.

Main Methods:

• Collected speech samples from 70 Parkinson's Disease patients.
• Extracted and concatenated diverse speech features from various speech tasks (/a/, /pa-ka-la/).
• Applied a proposed graph structure-based multi-label speech feature selection algorithm followed by a multi-label classifier.

Main Results:

• The proposed multi-label feature selection method demonstrated superior recognition performance compared to classical methods in most cases.
• The framework successfully recognized multiple PD subtypes simultaneously from speech data.
• Experimental results validate the effectiveness of the proposed approach for PD subtype recognition.

Conclusions:

• Multi-label learning from speech signals is a viable approach for Parkinson's Disease subtype recognition.
• The novel graph-based feature selection significantly enhances the accuracy of PD subtype identification.
• This speech-based method offers a promising tool for improving the diagnosis and management of Parkinson's Disease subtypes.