RELIABILITY OF BLOOD PRESSURE ESTIMATION FOR EXTREME VALUES

Objective: The use of a cuff for continuous blood pressure (BP) monitoring is limited by discomfort, interference with measurements and excessive time between measurements. The new cuffless methods based on pulse wave velocity seem to provide measurements with a correct accuracy. However, this accuracy may not maintained over the whole range of BP values. This study explores the accuracy of BP estimation using photoplethysmography (PPG) signals, in a wide range of intra-individual BP variations. Design and method: The cohort of the clinical trial (NCT05393401, approved by Human Research Review Board) included 18 healthy subjects (9F/9 M, age [19-60] years). Each subject undertook physical / mental activities and relaxation. The subject was monitored with electrocardiogram (ECG) and PPG sensors at the radial artery, at 1 kHz frequency. Signals were collected and synchronised by a custom-built multi-sensing platform. The Pulse Arrival Time was computed from ECG R-peaks and PPG signals for each heartbeat. Other inputs included subject’s age and body mass index, and pulse wave morphological metrics. Cuffless BP was estimated using the Adaboost regressor. In parallel, the subject’s BP was measured at every heartbeat with a medical-grade continuous measurement device. The model performance was quantified by the root mean square error (RMSE). Results: The measured systolic pressure ranged from 100 to 193 mmHg and the diastolic pressure between 49 and 106 mmHg. In this dataset, 84% of systolic values are within the interval [110;160] mmHg and are estimated with a RMSE of 13.4 mmHg. Similarly, 78% of diastolic values are within [60;90] mmHg and are estimated with a RMSE of 8.0 mmHg. For higher BP values, the estimation performances deteriorate with a RMSE of 22.0 mmHg for systolic and 14.5 mmHg for diastolic. For lower values, the RMSE are also worse with 26.8 mmHg for systolic and 11.6 mmHg for diastolic BP.Conclusions: The algorithms used to estimate BP from the pulse wave must be tested not only at small variations but also at extreme values to increase the robustness of the model. The experiments must also provide a significant proportion of high and low pressure values.