Oscillometry-Based Blood Pressure Estimation Using Convolutional Neural Networks

Blood pressure measurement is required to monitor the cardiovascular state of a person, and it is commonly conducted in a noninvasive way using oscillometry-based blood pressure monitors (BPM). Blood pressure can be estimated by analyzing the oscillometric waveform (OMW) in the BPM, and many methods have been examined to increase their estimation accuracy. In this study, we proposed a new method that enhances estimation accuracy and requires no external user information, such as age and gender, in the test phase. In the method, the entire OMW was considered as an input to reduce information loss via feature extraction, and convolutional neural networks were utilized to effectively analyze the high-dimensional input. Additionally, the proposed method included a novel ensemble method to further increase the estimation accuracy. The performance of the proposed method was evaluated and compared with other studies via subject-independent tests considering real situations in which it is difficult to obtain preliminary information on a test subject. Data from 64 subjects were used in the test. The mean absolute error of the proposed method was 3.12 and 3.98 mmHg for systolic and diastolic blood pressure, respectively, which was superior to those reported in other studies conducted in similar conditions. Individuals can measure their blood pressure with higher precision using the proposed method with improved estimation performance. This can aid in reducing the risk of cardiovascular diseases.