mSTEP: An Automated Algorithm for Blood Pressure Measurement from Korotkoff Sounds

Accurate blood pressure measurement is crucial for effective health monitoring, with oscillometric and auscultation methods being commonly employed. However, oscillometric measurements are population-dependent, and manual auscultation relies heavily on operator skill. This study introduces the modulated Short-Time Energy and finding the peaks (mSTEP) algorithm, designed to automate the tracking of systolic blood pressure (SBP) and diastolic blood pressure (DBP) from auscultatory signals, aiming to provide accurate BP measurements. The mSTEP algorithm was rigorously evaluated on a dataset comprising 350 recordings with manually annotated SBP and DBP for ground truth value. Implementation of the algorithm yielded successful identification of both SBP and DBP, with errors of -1.5 ± 6.6 mmHg and 3.8 ± 7.6 mmHg, respectively, with insignificant bias (p<0.0001). The agreement between mSTEP algorithm results and the reference measurement was strong and statistically significant, exhibiting correlation coefficients (r) of 0.94 (p<0.0001) for SBP and 0.79 (p<0.0001) for DBP. Bland-Altman plots further supported strong agreement, with confidence intervals of -14.5 mmHg to 11.4 mmHg for SBP and - 11.1 mmHg to 18.6 mmHg for DBP. According to the performance assessment, the algorithm received an 'A' grade in the BHS grading system and successfully met the accuracy requirements prescribed by AAMI/ANSI and ISO. The proposed algorithm potentially leads to the development of automated devices for accurate blood pressure measurements without depending on population or individual-level calibration coefficients.