Contactless Tracking of Blood Pressure Changes Using Cameras
No Thumbnail Available
Author
Bonnier, Guillaume
Aguet, Clémentine
Proença, Martin
Gimkiewicz, Christiane
Frosio, Monique
Springmann, Valerie
Saeedi, Sareh
Maamari, Nadim
Lemay, Mathieu
Braun, Fabian
DOI
10.1515/bmt-2025-1001
Abstract
Methods In a data collection campaign involving 150 healthy subjects, facial video and upper-arm-based PPG data were recorded simultaneously under controlled illumination conditions. Using these data, a CNN-based spatio-temporal deep learning model was trained to output rPPG signals from video. For a test set, 12 subjects were performing an exercise protocol (leg extensions) with concomitant cuff-based BP measurements. The rPPG signals generated from this test set were processed to estimate BP using CSEM’s clinically validated algorithm. Systolic and diastolic BP changes (ΔSBP and ΔDBP) were compared across subjects using mean ± standard deviation of differences between devices. BP tracking performance was also assessed using the concordance rate (CR), which measures the percentage of significant BP changes moving in the same direction on both devices.
Results For the 12 subjects (3 female; age: 34±7 yrs; BMI: 22.3±2.0 kg/m2) the agreement between the cuff-based and rPPGbased BP monitors was 1.2±8.7 mmHg for ΔSBP and 1.7± 7.9 mmHg for ΔDBP. Camera-based BP trending via rPPG showed a CR of 97.7% for SBP and 96.9% for DBP with an acceptance rate of 86±14%.
Conclusion CSEM's rPPG-based cuffless BP monitor accurately tracks BP changes compared to an oscillometric device under controlled conditions. If these promising results are validated in less controlled environments (e.g., natural lighting) and in large patient groups, it could revolutionize BP monitoring. Any smart device with a high-quality video camera could potentially measure changes in BP, greatly simplifying access to BP data.
Publication Reference
BMT 2025, Muttenz (Switzerland)
Year
2025-09-11