Mental Health Screening Using the Heart Rate Variability and Frontal Electroencephalography Features: A Machine...

Area of Science:

• Neuroscience and Computational Psychiatry
• Biomedical Engineering and Signal Processing

Background:

• Heart rate variability (HRV) reflects autonomic nervous system function and emotional regulation.
• Electroencephalography (EEG) measures cortical activity linked to psychopathology.
• Previous research utilized HRV and EEG, often with other wearable data, for psychiatric disorder (PT) classification, but few explored diverse diagnoses against healthy controls (HC).

Purpose of the Study:

• To identify key heart rate variability (HRV) and prefrontal electroencephalography (EEG) features for accurate classification of psychiatric disorders (PT) versus healthy controls (HC) using Support Vector Machine (SVM).
• To contribute to early, individual-level screening of PT and reduce the duration of untreated illness.

Main Methods:

• Simultaneous 5-minute photoplethysmography (PPG) for HRV and resting-state EEG data were collected from 182 participants (87 PT, 95 HC).
• Time- and frequency-domain HRV features and EEG power spectral density were analyzed.
• A feature selection process using one-way ANOVA F-value identified top N (1-22) HRV/EEG features for Gaussian radial basis function kernel SVM models.
• Leave-one-out cross-validation (LOOCV) assessed classification performance (balanced accuracy, AUROC) for PT vs. HC.

Main Results:

• The optimal SVM model with 13 features achieved a balanced accuracy of 0.76 and an area under the receiver operating characteristic curve (AUROC) of 0.78.
• Key features included HRV power spectral density (high, very low, low frequency bands, total power, SDNN) and frontal EEG alpha peak frequency and power spectral density.
• These top 13 features were consistently selected in over 90% of LOOCV iterations.

Conclusions:

• Combining HRV and prefrontal EEG features shows synergistic potential for machine learning-based mental health screening.
• Further research is needed to predict treatment response and guide therapeutic regimens using baseline physiological markers.