| This thesis investigates the detection of heart sounds from accelerometer signals found in implantable cardiac pacemakers. The detection algorithm creates an envelope of a filtered, signal-averaged accelerometer signal and then uses a threshold algorithm for detection. Three envelope techniques are investigated, a signal envelope, a Liang normalized energy envelogram, and an enhancement to the Liang normalized energy envelogram that incorporates a high pass filter. Two thresholding techniques are investigated, a threshold calculated using moving averages and a threshold calculated using a bimodal Kippler-Illingsworth technique. Each threshold technique was used with each envelope technique on nine different patient data sets to generate six different result sets. The Liang energy envelogram with the high pass filter enhancement performed best of the three envelope techniques using both the thresholding techniques, and it was also the only envelope technique found to produce good results when compared with earlier detection efforts in the literature. For these signals, it was also found that the moving average thresholding technique performed somewhat better than the bimodal Kippler-Illingsworth thresholding technique. |
