How accurate is Shen AI?

‍

Reliable health measurements start with reliable validation. Shen AI is continuously evaluated against medical reference devices to understand how closely its measurements match established clinical methods. These validation studies help us improve the SDK, measure progress over time, and give customers a transparent view of its performance.

This article summarizes the results of our latest clinical validation study. For the complete methodology, statistical analysis, and detailed performance tables, see the Clinical validation report.

‍

How do we validate Shen AI?

Our latest clinical validation study included 132 adult participants measured over two clinical visits. During each measurement, participants were recorded with a standard smartphone camera while medical reference devices measured the same vital signs at the same time. This allowed researchers to compare every Shen AI result with a trusted clinical reference.

The study evaluated four key measurements:

- Heart Rate
- Heart Rate Variability (HRV)
- Breathing Rate
- Blood Pressure

‍

Heart Rate

Heart Rate produced the strongest results in the study. Using a standard 60-second measurement, Shen AI achieved an average error of just 0.14 beats per minute with an almost perfect correlation (0.99) compared to ECG. Even after shortening the measurement to 30 seconds, performance remained very similar.

That's one of the reasons we were able to reduce the standard measurement time from 60 seconds to 30 seconds without compromising the user experience.

‍

Table 1: Summary of heart rate accuracy (Shen AI vs ECG)

‍

Heart Rate Variability

Heart Rate Variability is much more difficult to measure than Heart Rate because it looks at tiny differences between individual heartbeats rather than simply counting them.

Even so, Shen AI achieved a correlation of 0.95 with ECG measurements in the clinical study. Because HRV requires more data to produce stable results, it's always calculated using the full measurement window.

‍

Table 2: Heart rate variability (SDNN) accuracy

‍

Breathing Rate

Breathing is naturally slower than the heartbeat, so the SDK needs to observe more breathing cycles before it can produce a reliable estimate.

For that reason, breathing rate benefits from longer recordings. In our study, the average error for a 60-second measurement was 1.4 breaths per minute, while shorter measurements still produced results useful for many wellness applications.

‍

Table 3: Breathing rate accuracy

‍

Blood Pressure

Our clinical study evaluated both calibrated and non-calibrated blood pressure estimation. The results showed that individual calibration can improve agreement with traditional cuff measurements, and we continue to refine these algorithms through ongoing research and clinical validation.

‍

Table 4: Blood pressure accuracy

Note: MAE, RMSE, SD, and Bias are in mmHg; MAPE in %; r = Pearson correlation.

‍

In summary

Clinical validation is an ongoing part of how Shen AI is developed. Every study helps us better understand the strengths of the technology, improve existing algorithms, and validate new capabilities before they reach production.

The results presented here demonstrate strong agreement with established medical reference devices across the vital signs measured by Shen AI. For customers evaluating the SDK, they provide an independent benchmark of how the technology performs under controlled clinical conditions. If you want to test Shen AI’s accuracy yourself, follow the recommendations listed in this guide.

‍