1. Model Details

Basic info: Presage vitals by video analysis generates heart rate variability metrics (RMSSD, SDNN, Baevsky Stress Index) from a video of a subject. Output frequency: one measurement per 60 seconds.

Organization: Presage Technologies

Model date: 2026-04-01

Model version: 3.0.0-rc.11

Model type: A proprietary computer vision and signal processing pipeline extracts a photoplethysmogram (PPG) from video of a subject, from which NN intervals are derived. HRV metrics reported: RMSSD = sqrt(mean((NN_i+1 - NN_i)²)), the root mean square of successive NN differences (ms); SDNN = sqrt(mean((NN_i - mean(NN))²)), the standard deviation of NN intervals (ms); and Baevsky Stress Index = AMo / (2 × Mo × MxDMn), a measure of autonomic balance derived from the NN interval distribution.

License: The algorithm is currently proprietary, and licenses are granted with predefined agreement.

Contact: Questions can be sent to: support@presagetech.com

2. Intended Use

Model Uses

This HRV model was intended for use in analysis and non-diagnostic utility of HRV. It requires at least 60 consecutive seconds of unobstructed facial view acquired at 25+ fps. Only intended for participants with pulse rate 40-180 BPM.

Out-of-Scope Uses

As noted above. Not intended for diagnostic purposes, arrhythmia detection, or monitoring. Not intended for highly dynamic environments.

3. Validation

Reference Standard: Biopac MP160 3-lead ECG. R-peaks detected from ECG; NN intervals computed; HRV metrics derived from NN intervals over matched 60-second windows.

Comparison Methodology: HRV (RMSSD, SDNN, Baevsky) measurements from the camera-based system were compared against time-aligned reference measurements. Ground truth signals were checked for quality using labeled signal annotations; segments with poor signal quality were excluded from analysis.

4. Data Demographics

Category	Distribution
Total	119 subjects, 431 videos
Camera (videos)	Logitech C920: 118, e-con See3CAM CU27: 115, Samsung S24 Rear 24mm (tripod): 101, Samsung S24 Front 27mm (handheld): 97
Sex (subjects)	Female: 63, Male: 56
Age Group (subjects)	18-25: 38, 26-35: 35, 36-45: 18, 46-55: 15, 56-65: 10, 65+: 2
Fitzpatrick (subjects)	Type 1: 21, Type 2: 13, Type 3: 13, Type 4: 24, Type 5: 33, Type 6: 15
BMI (subjects)	Mean: 27.9, Range: 16.5–54.3, N=119
Lighting (videos)	Lamp: 137, Ring Light: 133, Window: 105

Reference standard: Biopac 3-lead ECG. Confidence values averaged across each 60-second period.

5. Data Provenance

Reference Instrumentation: Biopac research-grade physiological sensors: ECG (for HRV ground truth).

Camera Devices Tested: Samsung S24 Rear 24mm (tripod), Samsung S24 Front 27mm (handheld), e-con See3CAM CU27 (USB), Logitech C920 (tripod).

Average Camera Distance: Samsung S24/e-con/Logitech: 26", Samsung S24 handheld: 10"

Data Handling: All subject data is de-identified. Derived metrics and anonymized identifiers are retained. Data is securely stored with access restricted to trained researchers.

6. Factors

The HRV model requires face detection for photoplethysmographic signal extraction.

User: Face must be within angular detection parameters, visible and illuminated.
Environment: Developed on videos with various lighting and motion conditions.

These factors can affect model performance:

Lighting: Affects plethysmogram quality and thus HRV accuracy.
User: Skin tone and facial hair may impact signal quality.
Motion: High user motion degrades the plethysmographic signal.
Chewing gum can degrade performance.

Lighting Conditions Tested:

Ring Light: Direct, front-facing illumination providing consistent lighting on the subject's face.
Window: Natural sunlight entering through a window that illuminates the subject's face from the side rather than directly head-on. Lighting intensity varies depending on weather conditions and the time of day when the measurement is collected.
Lamp: A room lighting lamp aimed at the ceiling and not directed at the participant, to mimic living room-like environments.

Other factors:

Instrumentation: Camera parameter locking (exposure, gain, white balance) is critical for HRV accuracy.

7. Metrics

(at 80% Return Rate, Confidence >= 89)

Mean NN MAE: 3.40 ms
Mean NN RMSE: 5.06 ms
RMSSD MAE: 13.11 ms
RMSSD RMSE: 17.66 ms
SDNN MAE: 6.87 ms
SDNN RMSE: 10.42 ms

8. Quantitative Analysis

Computed vs Ground Truth at Confidence Thresholds

Bland-Altman Plot

Confidence Lookup Table

Confidence >=	Mean NN MAE (ms)	Mean NN RMSE (ms)	RMSSD MAE (ms)	RMSSD RMSE (ms)	SDNN MAE (ms)	SDNN RMSE (ms)	Return Rate (%)	N (samples)
0	9.15	21.96	26.57	45.67	16.54	32.21	100.0	17452
10	9.15	21.96	26.57	45.67	16.54	32.21	100.0	17452
20	7.22	15.97	23.02	37.72	13.87	26.09	97.8	17060
30	6.43	13.41	21.74	34.74	12.88	23.69	96.8	16897
40	5.82	11.81	20.16	31.33	11.76	21.02	95.4	16641
50	4.66	7.88	18.09	26.72	10.18	17.12	93.1	16244
60	4.33	7.00	17.15	24.68	9.55	15.61	92.1	16079
65	4.09	6.35	16.43	23.38	8.99	14.26	90.9	15860
70	3.98	5.94	15.85	21.97	8.55	13.13	89.7	15653
75	3.88	5.84	15.39	21.33	8.30	12.84	87.9	15335
80	3.76	5.67	14.83	20.21	8.01	12.38	86.1	15020
85	3.54	5.34	13.79	18.65	7.42	11.41	82.5	14394
89	3.40	5.06	13.11	17.66	6.87	10.42	80.0	13920
90	3.39	5.02	12.96	17.38	6.76	10.23	79.1	13796
95	3.04	4.18	11.54	15.13	5.81	8.23	70.3	12271

Performance

(at 80% Return Rate, Confidence >= 89)

By Camera Type

Camera Type	N (samples)	Return Rate (%)	Mean NN MAE (ms)	Mean NN RMSE (ms)	RMSSD MAE (ms)	RMSSD RMSE (ms)	SDNN MAE (ms)	SDNN RMSE (ms)
Samsung S24 Rear 24mm (tripod)	3934	90.4	3.59	5.47	11.93	15.89	6.27	8.52
Samsung S24 Front 27mm (handheld)	1625	56.2	3.05	4.80	14.52	19.52	7.71	12.72
e-con See3CAM CU27	4094	79.8	3.53	4.99	14.00	19.24	7.61	12.00
Logitech C920	4267	84.1	3.36	4.93	12.37	16.16	6.14	8.42

By Fitzpatrick Skin Type

Fitzpatrick	N (samples)	Return Rate (%)	Mean NN MAE (ms)	Mean NN RMSE (ms)	RMSSD MAE (ms)	RMSSD RMSE (ms)	SDNN MAE (ms)	SDNN RMSE (ms)
Type I	2472	94.9	2.94	3.90	8.18	10.66	3.53	4.49
Type II	1748	90.1	2.57	4.35	9.19	11.18	5.19	8.89
Type III	1724	85.9	4.92	7.79	11.43	17.07	6.65	10.70
Type IV	3300	92.1	3.20	4.66	12.68	17.09	6.79	9.38
Type V	3870	75.2	3.51	4.79	17.55	22.22	9.32	13.43
Type VI	806	37.1	4.00	5.58	19.50	22.87	9.01	11.93
Fitzpatrick Type VI: Breakdown by Lighting Condition
Ring Light	337	46.7	4.59	5.91	15.27	18.53	9.05	11.46
Window	175	27.8	5.65	8.01	32.06	33.18	9.17	10.29
Lamp	248	33.7	2.32	2.76	19.26	22.55	8.00	13.50

Note: Types I-V are composites across all lighting conditions (Ring Light, Window, Lamp).

By Sex

Sex	N (samples)	Return Rate (%)	Mean NN MAE (ms)	Mean NN RMSE (ms)	RMSSD MAE (ms)	RMSSD RMSE (ms)	SDNN MAE (ms)	SDNN RMSE (ms)
Male	7233	85.3	3.42	5.34	11.51	15.27	5.70	8.36
Female	6687	74.5	3.38	4.71	14.95	20.05	8.22	12.37

By Age Group

Age Group	N (samples)	Return Rate (%)	Mean NN MAE (ms)	Mean NN RMSE (ms)	RMSSD MAE (ms)	RMSSD RMSE (ms)	SDNN MAE (ms)	SDNN RMSE (ms)
18-25	4655	81.5	4.28	6.39	11.76	16.48	6.83	10.11
26-35	3817	77.2	3.58	4.84	9.93	14.80	5.87	10.93
36-45	1911	76.7	3.79	5.23	15.18	18.23	6.70	8.24
46-55	2082	90.6	1.69	2.29	18.64	22.29	8.43	11.53
56-65	1308	82.4	2.12	2.96	12.91	17.22	7.09	10.51
65+	47	16.4	9.23	10.48	37.96	38.52	11.77	12.63

By Lighting Type

Lighting	N (samples)	Return Rate (%)	Mean NN MAE (ms)	Mean NN RMSE (ms)	RMSSD MAE (ms)	RMSSD RMSE (ms)	SDNN MAE (ms)	SDNN RMSE (ms)
Lamp	4300	77.3	3.36	4.46	12.68	16.90	7.21	11.36
Ring Light	4707	83.7	3.17	4.74	14.41	19.35	7.59	11.48
Window	2771	69.1	4.40	7.12	11.79	17.00	6.07	8.96

9. Fairness & Equity

Bias Assessment Methodology: Performance is stratified by Fitzpatrick skin type (I-VI), sex, camera type, and age group. Per-group metrics and Confidence averages are reported in the Quantitative Analysis tables above.

10. Ethical Considerations

As a remote sensing device, the risks posed to the subjects in the trial are minimal, including the association of each subject with corresponding biometric data. Mitigation of these risks include de-identifying all subject data, including videos, prior to saving it. Additionally, all data is securely stored with access to a select number of trained researchers.

The model is not intended for human life-critical decisions, diagnostics or prognostication.

11. Limitations and Tradeoffs

Requires at least 60 seconds of uninterrupted facial view
Works best in stationary conditions
HRV metrics are sensitive to motion artifacts
Currently only measures a single subject at a time

Presage HRV Model Card