Automated Parkinson's disease recognition based on statistical pooling method using acoustic features.

Area of Science:

• Neurology
• Machine Learning
• Biomedical Signal Processing

Background:

• Parkinson's disease (PD) is a prevalent neurological disorder impacting motor functions like speech and balance.
• Current diagnostic methods often rely on clinical assessments, with machine learning exploring acoustic, handwriting, and gait data.
• Automated detection of PD is crucial for timely intervention and management.

Purpose of the Study:

• To propose and evaluate a novel statistical pooling method for Parkinson's disease recognition using vowel acoustic features.
• To enhance feature representation and identify the most significant discriminators for improved classification accuracy.
• To compare the effectiveness of Support Vector Machine (SVM) and K Nearest Neighbor (KNN) algorithms with the proposed method.

Main Methods:

• A statistical pooling method was applied to augment features from a Parkinson's disease acoustic dataset (vowel features).
• The ReliefF algorithm was employed to select the most weighted features from the expanded feature set.
• Classification was performed using Support Vector Machine (SVM) and K Nearest Neighbor (KNN) algorithms on the selected features.

Main Results:

• The proposed method achieved high success rates: 91.25% with SVM and 91.23% with KNN.
• Statistical pooling effectively generated new, informative features from the acoustic data.
• Feature selection using ReliefF identified the most significant discriminators for PD detection.

Conclusions:

• The proposed statistical pooling and feature selection method demonstrates significant success in recognizing Parkinson's disease from acoustic data.
• The approach offers a valuable contribution to automated PD diagnosis, outperforming selected state-of-the-art methods.
• This technique provides a robust framework for leveraging acoustic features in the early detection of Parkinson's disease.