Cuffless Non-invasive Blood Pressure Measurement Using CNN-LSTM Model: A Correlation Study

Abstract

Cardiovascular disease is a major concern for people all around the world and still remains as the main cause of death worldwide. Blood pressure has been identified as the most important risk factor. Having the ability to acquire continuous monitoring on this biological parameter plays a significant role in reducing the risk of getting cardiac disease. Many studies conducted utilize two biosignals and features manually extracted from signals as input to the model. However, these methods increase the computational complexity in the pre-processing stage as it involves signal synchronization, and the model performance is highly dependent on the selection of features. The main objective of this study is to build a hybrid convolutional neural network combined with Long-Short Term Memory (CNN-LSTM) model to estimate blood pressure from PPG signals, which eliminates the need for manual feature extraction. Correlation study is performed to evaluate the performance of the model, and it gives a direct visualization of the model’s performance in percentage. This research compared the correlation performance between MIMIC-II dataset, UKM dataset, and PPG-BP dataset using the CNN-LSTM model to estimate blood pressure from PPG signals. The results show that the UKM dataset performs the best, having the highest overall correlation at 0.53 for systolic blood pressure, and 0.29 for diastolic blood pressure. The model trained with this dataset is suitable to estimate systolic blood pressure ranging from 141 to 150mmHg, and diastolic blood pressure ranging 81 to 90 mmHg. In conclusion, among the three datasets, UKM dataset is the most suitable dataset to be used as the input of the CNN-LSTM model to perform cuffless blood pressure measurement with PPG signals.