# Assessing Mental Health and Emotional States by Using Smartphone Photoplethysmography–Based Digital Pulse Waveform Analysis: Cross-Sectional Observational Study

## TL;DR

This study shows that smartphone-based pulse analysis can detect links between pulse features and mental health, especially for negative emotions, but with limited accuracy.

## Contribution

The study introduces a novel method for assessing mental health using smartphone photoplethysmography and identifies specific pulse features linked to psychological states.

## Key findings

- Negative psychological states like depression and anxiety are significantly associated with time- and curvature-domain pulse features.
- Random forest models showed modest predictive performance for negative mental health outcomes but weaker results for positive affect.
- Smartphone-derived pulse features showed acceptable agreement with oximeter measurements, though signal quality varied.

## Abstract

Pulse characteristics are well-established biomarkers of physical health; however, their relevance to psychological well-being remains insufficiently explored. A key barrier is the difficulty of acquiring pulse recordings and blood pressure measurements of adequate quality outside clinical or laboratory settings by using accessible measurement approaches.

This study aimed to examine the feasibility of using smartphone photoplethysmography to extract fingertip pulse-waveform features and to evaluate their associations with psychological measures. It further aimed to systematically compare time-, curvature-, and frequency-domain pulse-waveform features in relation to psychological variables.

A total of 127 students and university employees in Shenzhen, China, were recruited. Participants recorded repeated 4-minute fingertip videos by using a custom smartphone app while a fingertip oximeter simultaneously acquired reference pulse signals. Smartphone videos were converted into photoplethysmography signals, segmented into beat-to-beat intervals, and summarized into time-, curvature-, and frequency-domain features, with normalization for heart rate and stature. Psychological well-being and mental health were assessed using the Satisfaction With Life Scale, Subjective Vitality Scale, Positive and Negative Affect Schedule, Patient Health Questionnaire-9, Generalized Anxiety Disorder-7, and the Self-Assessment Manikin. Associations between pulse-waveform features and psychological measures were examined using univariate regression with participant-level aggregation and cluster-robust standard errors. Random forest models evaluated multivariate predictive performance by using participant-level cross-validation. Agreement between smartphone-derived and oximeter-derived waveform features was assessed using Bland–Altman analysis.

Correlation analyses revealed strong within-domain associations among time-, curvature-, and frequency-domain pulse-waveform features, with comparatively weaker cross-domain correlations. A correlation-based feature-selection procedure reduced multicollinearity and yielded a final set of 7 features: estimated reflection index, crest time (CT), the third curvature minimum (F/A), the fourth curvature minimum (H/A), the first power spectrum density component, the baseline of Fourier decomposition (V0), and systolic blood pressure. Univariate regression analyses indicated that negative psychological states were primarily associated with time- and curvature-domain features. Depressive symptoms were significantly related to F/A, V0, and the estimated reflection index. Anxiety showed an association with F/A, and negative affect was associated with CT and F/A. In contrast, positive affect measures showed fewer and weaker associations. Valence was related to F/A and H/A, whereas arousal was associated with CT and H/A. Random forest models demonstrated statistically significant but modest predictive performance for negative mental health outcomes, with weaker performance for positive affect. Bland–Altman analyses indicated minimal systematic bias for outcomes with significant predictive correlations. Comparisons with an oximeter showed significant correlations and acceptable agreement, with time-domain features demonstrating greater robustness than reflection-based metrics.

Smartphone-based photoplethysmography can capture pulse-waveform features associated with psychological measures, particularly negative psychological states. However, predictive performance remains limited, and variability in signal quality from user-operated recordings poses a practical challenge.