Blood pressure measurements with the OptiBP smartphone app validated against reference auscultatory measurements

Abstract

Mobile health diagnostics have been shown to be effective and scalable for chronic disease detection and management. By maximizing the smartphones' optics and computational power, they could allow assessment of physiological information from the morphology of pulse waves and thus estimate cuffless blood pressure (BP). We trained the parameters of an existing pulse wave analysis algorithm (oBPM), previously validated in anaesthesia on pulse oximeter signals, by collecting optical signals from 51 patients fingertips via a smartphone while simultaneously acquiring BP measurements through an arterial catheter. We then compared smartphone-based measurements obtained on 50 participants in an ambulatory setting via the OptiBP app against simultaneously acquired auscultatory systolic blood pressure (SBP), diastolic blood pressure (DBP) and mean blood pressure (MBP) measurements. Patients were normotensive (70.0% for SBP versus 61.4% for DBP), hypertensive (17.1% vs. 13.6%) or hypotensive (12.9% vs. 25.0%). The difference in BP (mean ± standard deviation) between both methods were within the ISO 81,060-2:2018 standard for SBP (- 0.7 ± 7.7 mmHg), DBP (- 0.4 ± 4.5 mmHg) and MBP (- 0.6 ± 5.2 mmHg). These results demonstrate that BP can be measured with accuracy at the finger using the OptiBP smartphone app. This may become an important tool to detect hypertension in various settings, for example in low-income countries, where the availability of smartphones is high but access to health care is low.

Conflict of interest statement

Patrick Schoettker is an advisor to Biospectal. Martin Proença, Guillaume Bonnier, Alia Lemkaddem and Mathieu Lemay are with CSEM, the owner of the oBPM technology and assignee of the oBPM patent application (WO2016138965A1), of which Martin Proença and Mathieu Lemay are inventors. Jean Degott, Gregory Hofmann, Raoul Schorer, Arlene Wuerzner, Michel Burnier and Gregoire Wuerzner have no competing interest. Urvan Christen and Jean-François Knebel are working for Biospectal SA. Innosuisse—Swiss Innovation Agency, Project no. 32688.1 IP-ICT had no role in study design, data collection nor analysis, in the writing of the report nor in the decision to submit the paper for publication.

Figures

Figure 1

OptiBP application utilizes image data generated from volumetric blood flow changes via light passing through the fingertip, reflecting off of the tissue, and then passing to the phone camera's image sensor.

Figure 2

Algorithm description, parameter training, and calibration. (A) Working principle of the oBPM (optical blood pressure monitoring) algorithm. The oBPM algorithm automatically identifies all individual pulses in the PPG signal and ensemble averages them. Pulses with un-physiological morphologies (red dots) are identified and assigned low weights in the ensemble averaging procedure, whereas the remaining pulses (green dots) are assigned a stronger influence. The resulting ensemble average waveform is fed to a filter bank of time-derivative filters, allowing a decomposition of the waveform at various time resolutions. From their outputs, a set of features x characterizing the morphology of the waveform is obtained and nonlinearly combined using a pre-trained set of parameters θ^ (see (B) panel of the figure). The result is an uncalibrated BP value, BPoBPMUncal. The final oBPM-derived BP estimate (BPoBPM) is obtained after application of the previously trained corrective calibration offset β^. (B) Training of the parameters of the oBPM algorithm. The parameters were trained using the data acquired in the operating room. Significant BP changes (ΔBPInv≥±20%) between successive recordings were identified in the arterial line measurements. Their corresponding oBPM-derived BP changes (ΔBPoBPM) were then calculated to be compared. The set of oBPM parameters θ was optimized by minimizing the cohort-wise error between ΔBPoBPM and ΔBPInv in the least-square sense. In the figure, Nk is the number of significant BP changes found for patient k, and F is the non-linear oBPM model mapping the features x to BP values using the parameters θ. (C) Illustration of the calibration procedure. The calibration consists in the addition of a per-patient corrective offset β^ to the uncalibrated oBPM-derived BP estimate BPoBPMUncal for systolic, diastolic and mean BP individually. It is illustrated here with numerical values for ease of understanding. During the calibration measurement, the corrective offset β^ is calculated. Applying the calibration to the following test measurements consists of the addition of β^ to the uncalibrated BP estimate outputted by oBPM.

Figure 3

CONSORT Flow Chart of signals used for validation. 1SAP > 12 mmHg or DAP > 8 mmHg.

Figure 4

Standardized Bland–Altman scatterplots depicting the agreement between the OptiBP smartphone app systolic estimates assessed by the oBPM algorithm (SBPoBPM) and the auscultatory-derived reference systolic BP measurements (SBPAusc).

Figure 5

Standardized Bland–Altman scatterplots depicting the agreement between the OptiBP smartphone app diastolic estimates assessed by the oBPM algorithm (DBPoBPM) and the auscultatory-derived reference systolic BP measurements (DBPAusc).

Figure 6

Standardized Bland–Altman scatterplots depicting the agreement between the OptiBP smartphone app mean BP estimates assessed by the oBPM algorithm (MBPoBPM) and the auscultatory-derived reference systolic BP measurements (MBPAusc).