Algorithm development for artefact detection in PPG signal

Photoplethysmography (PPG) is a generally utilized non-invasive optical sensing technology to measure the cardiovascular parameters such as oxygen saturation, blood pressure, heart rate and etc. The changes in tissue and blood volume can be measured by PPG based on variations of emitted light on tissues in optical absorption and scattering. However, the PPG signal can be corrupted by artefacts causing difficulties in analysis of physiological parameters. This dissertation aims to detect the artefact contained in PPG signal. The proposed framework includes two complementary classification approach. The first one is based on time domain and correlation analysis which consists of two processing stages, cycle separation and similarity evaluation. The systolic peaks are extracted using the sum of the slopes to achieve efficient cycle separating and further processed by an improved correlation computation method. The result can be obtained by thresholding the evaluation output. The second method is based on support vector machine (SVM) model. The features used in SVM training are extracted by modelling the signal with fundamental and harmonics verified by a discrete-time Fourier transform (DTFT) based algorithm for Fourier coefficients estimation. Bayesian optimization is used to obtain the optimal hyperparameter. The results during the iterations are verified using 10-fold cross-validation. In addition, essential discriminative information is added to enhance the performance of the classifier. The classification methods, one of which is the time domain approach based on correlation measurement, identifies the artefacts with 95.85% accuracy. And the other classification system to analyze the PPG on a beat-to-beat basis using SVM can distinguish between the normal PPG and artefacts with 95.39% accuracy.Master of Science (Signal Processing