Blood Pressure Prediction with Long-term Recurrent Convolutional Networks Leveraging In-house Data and Mimic Data

Accurate blood pressure (BP) prediction is pivotal for effective cardiovascular health management. This study introduces a deep learning model that utilizes electrocardiogram (ECG) and photoplethysmogram (PPG) signals for effective BP predictions. Data from 16 subjects were collected. Utilizing a long-term recurrent convolutional network, the mean absolute error (MAE) achieved was 10.7±10.6 mmHg for systolic blood pressure (SBP) and 8.5±8.7 mmHg for diastolic blood pressure (DBP). These results underscore the constraints imposed by the small dataset size. Recognizing the necessity for additional data for optimal model training, the MIMIC dataset was leveraged, encompassing data from 900 subjects. The model exhibited predictive accuracy, yielding MAE values of 4.2 ± 5.6 mmHg for SBP and 2.7± 3.46 mmHg for DBP. Conforming to the benchmarks set by the AAMI, these figures serve as a solid affirmation of the model's precision. These results underscore the model's potential for practical clinical applications, offering a non-invasive continuous and efficient means of predicting BP. Such capabilities can enable early detection of cardiovascular conditions, enhancing patient care and outcomes.