Smartphone Camera Measurement of Breathing-Related Heart Rate Variability Compared With ECG in Healthy Adults

Design

Single-visit, single-group, within-participant method-comparison study. Each participant completes nine breathing conditions:

• Free breathing without pacing, 5 minutes (always first)
• Paced breathing at 4.0, 4.5, 5.0, 5.5, 6.5 and 7.0 breaths/min, 3 minutes each, and at 6.0 breaths/min, 5 minutes (participant-specific randomized order)
• Paced breathing at 15.0 breaths/min, 3 minutes (always last)

Pacing uses audio cues with an inhalation:exhalation time ratio of 4:6 at a comfortable depth, without breath holding. Conditions are separated by rests of at least 2 minutes, during which the fingertip is lifted and the cameras and flashlight are stopped. Planned recording time is 31 minutes plus 16 minutes of rest (47 minutes). Consent, sensor setup, at least 5 minutes of seated rest before recording, countdowns, data checks and a short questionnaire are additional; the visit lasts about 90 minutes. Participants are asked to avoid alcohol and strenuous exercise for 24 hours and caffeine, smoking and meals for 2 hours before recording. If these are not met, recording starts later or the visit is rescheduled; the times are recorded, and no participant is excluded on this basis after recording has started.

Order of conditions

The order of the seven slow rates is generated for each assignment number with a fixed random seed and stored before the participant's visit, balancing the position of each rate across participants as far as possible. The order is not changed after data are viewed. Free breathing is always first so that it is recorded before any paced breathing, and 15 breaths/min is always last so that faster breathing does not precede the slow rates. Differences between these two conditions and the slow rates may therefore include order and time effects; for 15 breaths/min these include carry-over from slow breathing and fatigue late in the session.

Measurements

A single iPhone 17 Pro running a research data-collection app uses the front camera (face) and the rear main camera (fingertip) simultaneously on a shared device clock. The fingertip is placed lightly over the rear camera lens while the flashlight beside the lens stays on and uncovered. No video from either camera is stored. For the face, the app continuously tracks regions on both cheeks and converts them into region-averaged signals. For every frame, the app saves the timestamp, the mean color values of the measurement regions, the detected face position and the camera settings, together with the extracted signals, beat candidates, device events and the time of every breathing cue. Further processing (beat detection, fiducial points and delay correction) follows an analysis specification frozen before the first participant is enrolled. A chest ECG (reference for beat timing) and a respiratory belt (reference for breathing rate and phase, and for adherence to pacing) record continuously from before the first condition until after the last. Start and end synchronization events are recorded to align the smartphone with the reference devices. Device model, app build and camera settings are fixed before the first participant is enrolled.

Analysis principles

• The participant is the unit of analysis. The conditions are repeated measurements within participants, not independent samples.
• Respiratory heart rate variability amplitude (A_pp, ms) is estimated separately for each sensor from inter-beat intervals (IBI) in the same time window by least-squares fitting of IBI(t) = mu + c*t + a*sin(phi(t)) + b*cos(phi(t)), where phi(t) is the phase of the guided breathing cycle, and A_pp = 2*sqrt(a^2 + b^2). A_pp is the peak-to-peak amplitude of the first harmonic at the guided frequency; it is not identical to every definition of respiratory sinus arrhythmia. No paced amplitude is assigned to free breathing.
• The signals saved by the app are processed offline with a pipeline whose code version, analysis windows, quality rules and failure definitions are fixed before evaluation data are analyzed. No enrolled participant is reassigned to a development set.
• All attempted recordings are retained. Loss of fingertip contact, loss of the face from the camera view or low signal quality counts as measurement failure and is not removed. Participants are not selected on the basis of sensor signal quality. ECG segments unsuitable as a reference are reported separately from sensor failures. A supplementary recording never replaces the original attempt: the primary analysis uses the original attempts, supplementary recordings are reported separately, and original attempts affected by contamination or technical problems are reported with the reason.
• Agreement is summarized by bias and repeated-measures 95% limits of agreement with participant-level confidence intervals; correlation alone is not used to judge agreement. Acceptable agreement and failure-rate limits are specified before evaluation data are analyzed; without such limits, results are reported as descriptive accuracy estimates without claims of equivalence or sufficiency.
• The primary analysis window is 60-180 s after the start of each 3-minute condition and 60-300 s after the start of each 5-minute condition (free breathing and 6.0 breaths/min); the first 60 seconds are excluded as a conservative settling period. Whether conditions could be shortened, by excluding less of the start or by analyzing a shorter recording, is examined in exploratory secondary analyses and does not change the primary windows.
• Analyses of alternative processing, shorter settling periods or recordings, and the breathing rate with maximal amplitude are labelled exploratory.
• Supplementary results, which are not separate outcome measures, include amplitude agreement expressed as absolute error in ms, results for each breathing rate, measurement failure among conditions with an eligible ECG reference, the proportion of valid recording time, RMSSD and SDNN agreement during the paced conditions, the movement of the detected face within each condition, and, among unmatched beats, the share that are displaced in time rather than absent or added.

Questionnaire

After all nine conditions, participants complete a short study-specific questionnaire once; it is not a validated instrument. Six items are rated from 0 to 10 for the average experience of the visit: difficulty keeping the face within the camera view, difficulty keeping the fingertip on the rear camera, fingertip discomfort, fingertip warmth or heat, being bothered by having the face filmed, and difficulty following the audio guide. For each method separately, participants then rate their satisfaction with the method as experienced that day (7-point item) and report any inconvenience. They are then handed the smartphone used in the session, may handle it as they like, and rate for each method how much they would want to use it alone for 5 minutes every night at home before bed (7-point items); they then choose the one method they would use (face, fingertip, or no preference), give the reason, and add suggestions. Face and fingertip items are presented in one of two orders (face first or fingertip first), assigned to each participant in advance from a random allocation table in blocks of four; 'no preference' is always the last option. The two difficulty ratings, their within-participant difference and the chosen method are outcome measures; the other items are reported descriptively. Items are analyzed separately without a total score, and missing responses are not imputed.

Sample size

The target is at least 20 evaluable participants, and up to 36 may be enrolled. Enrollment beyond 20 is intended to narrow the confidence intervals of the bias and limits of agreement and to allow for withdrawal or unusable reference recordings. The final number is determined by the recruitment and measurement schedule and budget, not by interim results: agreement is not analyzed before data collection ends, and checks during the study are limited to data completeness and recording integrity. All evaluable participants are analyzed; no subset is selected. The target of 20 reflects feasibility and is not derived from a power or precision calculation, because the variability and acceptable limits are not yet established.

Scope

Test-retest reliability, treatment effects of slow breathing, and confirmation of a physiological resonance frequency are outside the scope of this study.