An ECG-Based Blood Pressure Estimation Using U-Net auto-encoder and Random Forest Regressor

Measurements of Blood Pressure (BP) have become increasingly widespread in both clinical and private settings. In parallel, Electrocardiogram (ECG) monitors have also become increasingly prevalent. However, most ECG monitors currently available do not include the ability to estimate the value of BP. To address this gap, we have devised a novel BP estimation approach that relies solely on ECG signals. Our methodology involves a series of steps, including data filtering, and segmentation, and we thoroughly investigated the potential of using the auto-encoders of U-Net neural network, as an automatic feature extractor, followed by a regression algorithm in predicting the BP from the ECG. Using the MIMIC-II dataset, the model was trained. yielded mean absolute errors (MAE) of 6.0±4.49 mmHg (MAE±STD) and 2. 5±3.7 mmHg for Systolic Blood Pressure (SBP) and Diastolic Blood Pressure (DBP) respectively.