Photoplethysmography-based HRV analysis and machine learning for real-time stress quantification in mental health applications

Prolonged exposure to high-stress environments can lead to mental illnesses such as anxiety disorders, depression, and posttraumatic stress disorder. Here, a wearable device utilizing photoplethysmography (PPG) technology is developed to noninvasively measure physiological signals and analyze heart rate variability (HRV) parameters. Traditional normative HRV databases typically do not account for responses induced by specific stressors such as cognitive tasks. Therefore, machine learning is used to build a more dynamic stress assessment model. Machine learning can capture complex nonlinear relationships among HRV parameters during stress-inducing tasks, adapts to individual stress response variations, and provides real-time stress level predictions. Furthermore, machine learning models can integrate temporal patterns in HRV data to achieve nuanced stress level assessment. This study examines the feasibility of PPG signals and validates the developed stress model. The RR intervals derived from PPG signals were highly positively correlated with those from electrocardiography signals (correlation coefficient = 0.9920, R-squared = 0.9837); this confirms the usability of PPG signals for HRV analysis. The stress model is constructed via the open-source Swell dataset. In the experiments, participants complete the Depression Anxiety Stress Scales-21-Chinese (DASS-21-C) questionnaire to quantify levels of depression, anxiety, and stress over a week. Baseline and stress-state PPG data are collected, converted into HRV values, and input into the model for stress quantification. The Stroop test is used to elicit stress responses. After the experiment, the DASS-21-C stress scores were compared with the model's baseline, stress state, and combined scores. The highest correlation was observed between the model's baseline score and the DASS-21-C stress score (correlation coefficient = 0.92, R-squared = 0.8457), supporting the model's psychological significance in quantifying everyday stress. HRV parameter changes across experimental phases are discussed as well as sex differences in stress responses. In the future, this device may be applied in clinical scenarios for further validation and could be integrated with additional physiological indicators for broader application in daily health management and stress warning systems.