Feasibility study for detection of mental stress and depression using pulse rate variability metrics via various durations

Abstract

Background and ObjectiveIn modern society, people are exposed to various kinds of stress for a long time, which can lead to psychological or physical problems such as depression. This study aims to develop a pulse rate variability (PRV) detection model that can be used to assess mental states in real time, and to explore the differences and associations between acute mental stress and depression at the level of PRV features. MethodsPulse wave data were collected from 32 healthy college students during a mental arithmetic task and from 33 major depressive disorder patients in a resting state. 64 PRV features in the time, frequency, nonlinear and time–frequency domain from different data lengths (i.e., 1 min, 3 min and 5 min) were extracted, and the statistical analysis was used to investigate the significant difference of each feature under different time lengths. Three feature selection methods (i.e., mutual information, recursive feature elimination and random forest) combined with support vector machine were adopted to identify mental stress and depression. ResultsAfter feature selection, the highest accuracy for 1 min, 3 min, and 5 min were 90.77 %, 95.26 %, and 95.26 % when identifying mental stress, all 98.46 % when identifying depression, as well as 98.46 %, 98.46 %, and 100 % when differentiating between mental stress and depression. ConclusionsThis model provides a strong basis for effectively distinguishing between the three psychological states of health, mental stress, and depression, which can be applied to real-time monitoring and analysis of wearable devices.