Dual Prediction of Blood Pressure and Blood Glucose Level using PPG Signals: Explore Deep Learning Models through Comparative Study

Abstract

Monitoring blood pressure (BP) and blood glucose levels is vital for managing unremitting conditions such as hypertension and diabetes, which affect millions of individuals around the world. Conventional strategies for measuring BP, which depend exclusively on pulse amplitudes, often fail to provide exact readings, especially in patients with atherosclerosis or those who are obese, where pulse amplitudes may be frail or mutilated. Moreover, customary strategies for blood glucose estimation involve invasive strategies, causing discomfort. This research presents a novel approach to non-invasive dual prediction of BP and blood glucose levels utilizing photoplethysmogram (PPG) signals. Leveraging the power of LSTM network and XGBoostRegressor enables accurate and efficient forecast of both BP and blood glucose levels. LSTM is utilized to find time related patterns within PPG signal, while XGBoost enhances model performance by identifying the most relevant features for prediction. Then, the two pre-trained deep learning models (one for glucose prediction and another for blood pressure prediction) are loaded, and a test sample is reshaped to match the required input format for each model. These models are then used to make predictions on the reshaped sample. The results demonstrate promising performance, resulting in an RMS error of 8.014 mg/dL for blood glucose estimation and a Mean Absolute Error (MAE) of 25.336 mmHg for BP estimation. The model's performance reflects high potential for practical, non-invasive monitoring, offering a more comfortable and accessible alternative for patients requiring regular checking of both BP and glucose levels.