A CNN Approach To Predicting Non-Invasive Blood Pressure of Surgical Patients

The occurrence of intraoperative hypertension/ hypotension may cause danger to a patient. Therefore, the monitoring of blood pressure change during surgical operation is momentous for anesthetized patient undergoing surgical procedures. The invasive method of measuring arterial blood pressure (ABP) provides accurate and much information, but it requires evaluation by an anesthesiologist and with its long-term use, it also brings a higher risk of infection, and even some patients have pain and discomfort at the catheter placement site after surgery, so it is not suitable for all types of patients. On the other hand, non-invasive method of measuring cuff blood pressure provides little information because of non-continuous measurement. With the emerging of deep learning, there has been more and more discussion on whether non-invasive Photoplethysmography (PPG) can replace invasive ABP as an alternative to continuous BP measurement and even BP prediction. This paper aims to implement a deep learning model to predict SBP/DBP signal by PPG, PPG extended data and Electrocardiography (ECG) data. We trained the CNN-based architecture on preprocessed data including ECG, PPG, 1st and 2nd derivative PPG signal extracted from MIMIC-III(Medical Information Mart for Intensive Care III) waveform database matched subset to predict continuous ABP signals using real world data. The value of SBP and DBP are directly predicted to effectively estimate the accuracy of the predictions. The prediction results fulfill the grade A in British Hypertension Society (BHS) standard and most part of Association for the Advancement of Medical Instrumentation (AAMI)’s standard. The proposed model has been applied for CHANGHUA CHRISTIAN HOSPITAL’s IRB NO:191240. The results depict that our proposed model could effectively predict SBP/DBP signal and be deployed to the operating room realistically.