Beat by Beat Blood Pressure Classification Based on PPG and ECG Data Using a Deep Learning Framework

We developed a deep learning (DL) framework for classifying arterial blood pressure using photoplethysmogram (PPG) and electrocardiogram (ECG) signals. Physiological data, along with reference systolic (SBP) and diastolic (DBP) blood pressure, were collected from 38 subjects under varying physical exertion levels. After detecting fiducial points in the physiological signals, including R peaks in the ECG and peaks and valleys in the PPG, the entire data were segmented into individual heartbeats. The blood pressure associated with each beat was further classified into three categories: normal, prehypertension, and hypertension, based on corresponding DBP and SBP values. The classification of SBP and DBP was performed by a sequential convolutional neural network (CNN) model, specifically-designed to extract and classify patterns from 1D data. With each single pulse as input to the network, the model incorporated three Local Feature Learning Blocks, which consisted of progressively increasing filters, convolutional layers, and max pooling layers. The entire dataset was divided into training and testing sets, and the performance of the proposed framework was evaluated using accuracy, precision, and sensitivity metrics. The model demonstrated promising classification performance, achieving an accuracy of around 70% for both SBP and DBP on the three-class test set. Future studies in larger datasets are necessary to fullyevaluate the clinical value of the proposed DL model for noninvasive blood pressure classification.