Detecting Manic State of Bipolar Disorder Based on Support Vector Machine and Gaussian Mixture Model Using...

Area of Science:

• Computational psychiatry
• Machine learning in healthcare
• Speech signal processing

Background:

• Bipolar disorder (BD) diagnosis relies on clinical observation, which can be subjective.
• Objective, data-driven methods are needed for accurate and timely manic state detection.
• Speech analysis offers a promising, non-invasive approach for monitoring mood states.

Purpose of the Study:

• To compare the diagnostic accuracy of Support Vector Machine (SVM) and Gaussian Mixture Model (GMM) algorithms.
• To evaluate the performance of SVM and GMM in detecting manic states in bipolar disorder (BD) patients.
• To determine the optimal algorithm for single-patient versus multiple-patient manic state detection.

Main Methods:

• Recruited 21 hospitalized BD patients for speech data collection via smartphone.
• Extracted and preprocessed various speech features, including pitch, formants, MFCC, LPCC, and GFCC.
• Selected optimal features using between-class and within-class variance, then applied SVM and GMM for manic state detection.

Main Results:

• Linear Prediction Cepstral Coefficients (LPCC) exhibited superior discrimination efficiency.
• SVM achieved higher accuracy for single-patient manic state detection compared to GMM.
• GMM demonstrated superior accuracy for multiple-patient manic state detection compared to SVM.

Conclusions:

• SVM is suitable for individual manic state detection in BD.
• GMM is more effective for detecting manic states across multiple BD patients.
• Both SVM and GMM can aid clinicians and patients in diagnosis and mood monitoring.