A novel approach of Gaussian mixture model-based data compression of ECG and PPG signals for various cardiovascular diseases

To monitor cardiovascular functions, real-time ambulatory and telemedicine systems have limited storage capacity, necessitating the minimization of data quantity. This work proposes a compression method that preserves all salient aspects of Electrocardiogram (ECG) and Photoplethysmogram (PPG) for various cardiovascular diseases (CVDs) with fewer data bits. CVD classes—Right bundle branch block (RBBB), hypertension (Hyper), arrhythmia-atrial fibrillation (Ary-atrfab), dilated cardiomyopathy (DCM) along with one normal class with Lead-II ECG and PPG signals, extracted from the MIMIC-III dataset. Gaussian distribution functions were employed to fit these waves, reducing inter and intrabeat redundancy with fewer modeling parameters. PPG was fitted using six Gaussian functions, while ECG, with its greater morphological complexity, utilized three segments: P wave, QRS complex, and T wave, each fitted with two, six, and two Gaussian functions per single cardiac period, respectively. Also resilience was verified by MIT-BIH database and by real-world data. CR (20.956), PRD (3.131 %), PRDN (5.403 %), QS (6.960747), SNR (45.702 dB), RMSE (0.0116), CC (0.999464), and execution time (6.114 s) as performance values for the ECG. Likewise,PPG displays average values for CR (26.014), PRD (0.954 %), PRDN (2.691 %), QS (28.355), SNR (45.135 dB), RMSE (0.008), CC (0.999934), and execution time (5.293 s) for MIMIC-III. This methodology provides a simple, faster, effective, and low-computational complex data compression technique, which makes it a strong contender for use in remote health care or telemonitoring systems for CVD patients. Compared to existing compression techniques, this method efficiently manages both signals to store/transmit for diverse CVD patterns.