Phonocardiogram Noise Detection in Realistic Environments

One of the main obstacles that stalls the widespread use of phonocardiograms (PCGs) in modern day medicine is the various noise components they invariably contain. Al-though many advances have been made towards automated heart sound segmentation and heart pathology detection and classification, an efficient method for noise handling would come as a major aid for further development in this field, especially when it comes to working with PCGs collected in realistic environments such as hospitals and clinics. The use of a digital stethoscope to collect experimental samples of heart sounds in real-istic conditions results in highly noise contaminated PCGs. Besides the fact that these environments are often crowded, the very method of data gathering (the use of a trans-ducer for heart sound recording) contributes to the addition of background noise. Thus, only an adaptive algorithm that takes all these aspects into account will successfully increase signal-to-noise ratio. After a process of investigating and testing commonly used PCG noise reduction and noise detection techniques, we have selected a noise detection algorithm based on pe-riodicity measures to classify noisy and clean heart sound segments in PCGs collected in realistic environments. This thesis describes the process of selection, adaptation and testing of this algorithm. We will describe in detail the alterations performed in the previous implementation of this algorithm in order to enable processing of PCGs that contain high noise levels. The results achieved are satisfactory for the current stage of development and the conclusions taken clearly indicate a future line of work towards a robust noise detection algorithm that will aid signal processing of PCGs