Enhanced cuffless blood pressure estimation using ECG and PPG signals: A hybrid approach with Windkessel, ARIMA, and LSTM

Accurate blood pressure (BP) estimation is essential for the monitoring and management of cardiovascular diseases. This study presents a hybrid model for cuffless BP estimation using electrocardiogram (ECG) and photoplethysmogram (PPG) signals. The model incorporates features from time-domain, frequency-domain, and model-based approaches, including the Windkessel model, AutoRegressive Integrated Moving Average (ARIMA), and Long Short-Term Memory (LSTM) networks. To enhance performance, feature selection and reduction techniques such as Minimum Redundancy Maximum Relevance (MRMR) and autoencoders were employed. Additionally, model ensemble strategies, including average and weighted average modes, were utilized to combine the predictions of different models. The proposed method demonstrated superior performance with an RMSE of 2.98, MAE of 1.89, and R2 of 0.9326 for diastolic BP prediction, and an RMSE of 7.126, MAE of 4.684, and R2 of 0.9045 for systolic BP prediction. A total of 89,533 waveform records were used from the Multi-parameter Intelligent Monitoring in Intensive Care (MIMIC) II online waveform database.