| This thesis investigated the automated synchronous segmentation of cardiac cycles in phonocardiograms (PCGs) containing noise of varying degrees and types. Segmentation was performed using the Sliding Window Autocorrelation (SWA) algorithm. The SWA involved correlations between a heartbeat template and PCG, and relied on an estimate of heartbeat period. The objective of this thesis was to improve the accuracy and robustness of the SWA by addressing PCG variability in heart rate, heart sound intensity, and non-cardiac noise presence. Methods examined to enhance the SWA included an average heartbeat period estimate, an average heartbeat template estimate, a shifted heartbeat template competition, and data de-noising. An objective scoring method was introduced to evaluate segmentation accuracy. The results of this thesis indicated the SWA achieved correct segmentation, synchronous to within 50 ms of actual cardiac cycle boundaries, for 93.5% of the 2453 cardiac cycles in a dataset of 195 PCGs. Therefore, an improvement of 3.4% was achieved for the overall accuracy of the SWA. |
