| Abstract [eng] |
Cardiac arrhythmia is usually detected by using ECG, however it requires at least several electrodes attached to the body which is uncomfortable, especially during prolonged monitoring, e.g., day or more. In this thesis, a problem of cardiac arrhythmia, i.e., premature ventricular contractions (PVC) and atrial fibrillation (AF) detection by using photoplethysmogram (PPG) signals as an alternative to ECG is solved. Recording of PPG signals requires only a single electrode attached, e.g., to wrist, however PPG signals are susceptible to movement-induced noises (artifacts), which may cause false alarms. In order to address this problem following objectives were set: 1) to analyze existing methods for arrhythmia monitoring and bio-signal processing, 2) to develop and investigate PPG model capable of simulating various types of arrhythmia, 3) to develop and investigate method for the detection of PVC, 4) to develop and investigate methods for the detection of AF. Results of the research shows that: 1) currently the only PPG model of its kinds capable of simulating signals with various types of arrhythmia by using only the rhythm-based information was developed. The morphology of simulated PPGs corresponds to that of real signals. The model is intended for development and testing of the PPG-based arrhythmia detection methods, 2) a robust PPG-based PVC and bigeminy detection method employing heart-rhythm analysis, artifact detection and artificial neural network was developed, 3) a robust PPG-based AF detection method combining AF detection with PPG signal quality assessment, allowing to achieve high specificity even during low signal to noise ratios, was developed. |
