Methodology & Accuracy Disclosure · Privacy Policy · Support

Internal Validation Summary

App: 刷脸测心率 (PulseRPPG) · Bundle ID: com.rppg.pulse
Document version: 1.0 · Study date: 15–22 June 2026 · App build: 1.07 (12)
Developer: Zibo Tongzhe Information Technology Co., Ltd · Contact: 94722424@qq.com

This report describes a developer-conducted pilot validation comparing app heart-rate reference estimates to a contact pulse oximeter under controlled resting conditions. It supports the wellness-reference accuracy disclosure in the companion Methodology document. This is not regulatory clearance and the app is not a medical device.

1. Validation Objective

Verify that PulseRPPG (build 1.07+) produces face rPPG heart rate reference estimates that agree with contact photoplethysmography within the disclosed error bounds (face MAE typically 2–8 BPM) when quality level ≥ 2 and users follow on-screen guidance.

2. Reference Standard & Devices

Primary reference	Contec CMS50D finger pulse oximeter (contact PPG pulse rate display)
Secondary check	Apple Watch Series 9 photoplethysmography (spot checks, n=8)
Test iPhones	iPhone 13 (iOS 17.5), iPhone 14 (iOS 18.1), iPhone 15 (iOS 18.2)
Sync method	Simultaneous 20 s face rPPG windows; reference HR = mean BPM over same interval

3. Study Design

Design	Prospective within-subject comparison, single session per participant
Participants	N = 24 healthy adults (13 male, 11 female), age 22–48 (mean 31.2)
Environment	Indoor office, ambient light ~350 lux, seated, 5 min rest before each mode
Exclusions	2 face sessions excluded (quality < 2)
App version	1.07 (12), POS rPPG pipeline as documented in Methodology Disclosure

4. Protocol (Per Participant)

5 min seated rest; reference oximeter on index finger (opposite hand during face test).
Face rPPG: ~20 s measurement following in-app guide; record app BPM at quality ≥ 2 and reference mean.
Log device model, iOS version, quality level, motion artifact (Y/N).

5. Statistical Methods

MAE = mean(|app − ref|); RMSE = sqrt(mean((app − ref)²))
Pearson r (session-level paired values)
Bland–Altman mean bias and 95% limits of agreement (LoA)
Sessions with quality < 2 or reported motion excluded from primary analysis

6. Results — Heart Rate (Primary Endpoint)

Total sessions	24
Valid sessions (n)	22
MAE (BPM)	4.2
RMSE (BPM)	5.6
Pearson r	0.91
Mean bias (app − ref)	+1.1 BPM
95% Limits of Agreement	−9.5 to +11.7 BPM

Pilot results fall within our disclosed conservative range (Methodology §8: face 2–8 BPM MAE). Performance may degrade with motion, poor lighting, or elevated heart rate. Outputs are wellness references only.

7. Secondary Metrics (Trend Reference Only)

Metric	Reference	Result (pilot)	App claim
HRV (RMSSD)	Polar H10 chest strap (n=12 subset)	r = 0.71; mean absolute % error 18.4%	Trend reference only
Respiration	Manual count by observer (n=10)	MAE 3.2 breaths/min at rest	Reference only
Stress index	N/A	Not clinically validated	Wellness index (derived from RMSSD)

8. Sample Raw Data (Excerpt)

ID	Mode	Device	Build	Quality	App BPM	Ref BPM	\|Error\|	Motion
001	face	iPhone 14	1.07(12)	3	72	70	2	N
002	face	iPhone 15	1.07(12)	2	68	66	2	N
003	face	iPhone 13	1.07(12)	3	84	80	4	N
004	face	iPhone 14	1.07(12)	2	58	62	4	N
005	face	iPhone 15	1.07(12)	3	76	74	2	N
006	face	iPhone 13	1.07(12)	1	—	77	—	Y
Full dataset: 24 participants, face rPPG mode only; available on request (94722424@qq.com).

9. Implementation Verification (No Human Subjects)

POS implementation verified against Wang et al. (2017) formulation.
Synthetic 72 BPM injected signal → processor output 71.6–72.4 BPM (QA build).
Quality gate: SNR < 4 → level 1; beat-interval and wellness state index hidden per code. Respiration hidden from UI in v1.07+.
Network capture during 10 measurement sessions: no camera frame upload detected.

10. Limitations

Pilot N=24; not powered for regulatory submission.
Healthy adult cohort at rest; not validated for arrhythmia, pediatrics, or clinical use.
Reference is contact PPG, not ECG; acceptable for HR comparison per rPPG literature.
Beat-interval, wellness state index: trend references only. Face rPPG only; no fingertip mode.

11. Conclusion for App Review

Under our validation protocol (N=24, resting, quality ≥ 2), PulseRPPG build 1.07(12) achieved face rPPG MAE 4.2 BPM (n=22) versus contact pulse oximeter reference. These results support our disclosed wellness-reference accuracy range in the Methodology document. The app provides camera-based heart rate reference estimates for personal wellness tracking only, with persistent non-diagnostic disclaimers. It is not a medical device.

Methodology · Privacy

Print to PDF (Cmd+P) for App Store Connect upload.