Analysis of heart rate variability and subtle ECG changes based on machine learning for objective assessment of the psychological state of military personnel

TL;DR

This study explores using ECG and machine learning to assess the psychological state of military personnel, finding strong correlations with anxiety levels.

Contribution

The paper introduces a novel method for predicting psychological assessments using ECG and HRV data combined with machine learning.

Findings

ECG and HRV features showed strong correlations with the Beck Anxiety Scale and Preliminary Psychological Conclusion.

Machine learning models achieved R2 scores of 0.520 (training) and 0.359 (test) for predicting anxiety levels.

The approach demonstrates potential for objective psychological assessment using cardiovascular data.

Abstract

The implementation of objective methods for rapid assessment of the psychological and physiological readiness of military personnel is an extremely relevant task. The cardiovascular system acts as a “mirror” of functional and psychological state. The most common and accessible method for the objective study of the cardiovascular system remains electrocardiography (ECG). This study aims to develop a technology for objective monitoring of the psycho-emotional state and overall functional condition of personnel in the Ukrainian Defense Forces using miniature ECG devices and in-depth analysis of ECG signals with artificial intelligence.

Using an innovative ECG device, 90 servicemen, average age of 38 years, undergoing sanatorium treatment and rehabilitation at the Central Military Clinical Sanatorium “Khmilnyk” were examined. The examination was conducted on the first or second day after the start of sanatorium treatment. ECG and HRV analysis were performed using our previously developed Universal Scoring System. The results of ECG analysis from limb leads in 6 leads were compared with 4 well-known psychological self-assessment methods: Beck Anxiety Scale, PCL-5, PHQ-9, as well as a formalized psychologist’s conclusion. Correlation analysis and Sequential Feature Selector were used.

Forty ECG/HRV features were selected for each of the four psychological methods to find the maximum R2 metric. The highest number of reliable correlations between ECG and HRV parameters and psychological tests was found for the Beck Anxiety Scale. The same can be said when using feature selection via machine learning. The cross-validated R2 scores for the training and test sets in the case of the Beck Anxiety Scale were 0.520/0.359, respectively. Similar results were obtained for the Preliminary Psychological.

The study’s results demonstrate the potential for significant prediction of routine psychological assessment outcomes based on in-depth analysis of ECG and HRV, especially regarding the Beck Anxiety Scale and Preliminary Psychological Conclusion.

Figures