Aims: Abnormal heart rate (HR), respiration rate (RR), and blood pressure (BP) are the primary-most indicators of physiological instability, including chronic cardiopulmonary diseases. The study aimed to develop a deep learning framework, CP-Net, for simultaneously monitoring HR, RR, systolic and diastolic BP (SBP and DBP) from photoplethysmography (PPG). The employment of PPG was done mainly to extract all these parameters unobtrusively with a reduced number of attachments to the patient's body suitable for long-term continuous monitoring even in ambulatory and home-environment.
Methods: The proposed deep learning framework extracted all four parameters simultaneously without extracting any feature manually from PPG. Moreover, gated and long-short-term recurrent convolutional networks were incorporated for faster and more accurate estimation of those parameters. The performance of CP-Net was tested on a publicly available MIMIC-III database which comprises simultaneously recorded PPG, respiration, electrocardiogram (ECG), and arterial blood pressure signals collected from intensive care unit patients. Signals with missing, unreliable, and discontinued values were excluded from the dataset. The entire dataset was divided into training-testing subsets using a 5-fold cross-validation technique. The model's performance was evaluated by calculating the normalized mean absolute error (NMAE) and normalized root mean square error (NRMSE).
Results: Out of 38 subjects, signals collected from 31 subjects were selected in this study for further assessment. The model yielded NMAE of 0.0714 breath rate per minute (brpm), 0.0619 beats per minute (bpm), 0.1620 mmHg,and 0.0321 mmHg, and NRMSE of 0.1054, 0.071, 0.205, 0.039 for RR, HR,SBP, and DBP respectively and the overall loss was 0.0597 along with calculated errors for all four parameters.
Conclusion: The results exhibited that there was significantly less difference between actual and predicted values, which reflected the effectiveness of the proposed framework for the simultaneous measurement of all four vital parameters from PPG.