People with serious heart conditions have had their expected life span extended considerably with the development of the prosthetic heart valve especially with the great strides made in valve design. Even though the designs are extremely reliable, the valves are mechanical and operating continuously over a long period; therefore structural failures can occur due to fatigue. In this paper acoustical signal processing techniques developed to process noisy heart valve sounds measured by a sensitive, surface contact microphone are discussed. Measuring heart sounds noninvasively in a noisy environment puts more demands on the signal processing to extract the desired signals from the noise. Heart valve sounds are short‐duration (10–20 ms) transients and therefore nonstationary, requiring more sophisticated processing algorithms to achieve the desired signal‐to‐noise ratios. In this paper the preclassification signal processing is concentrated on exclusively. That is, the signal processing operations performed on the heart valve sounds prior to classification are discussed—a subject that will be developed in a future paper.
Efforts are concentrated on the sounds corresponding to the heart valve opening cycle. Valve opening and closing acoustics present additional information about the outlet strut condition—the structural component implicated in valve failure. The importance of the opening sound for single leg separation detection/classification is based on the fact that as the valve opens, the disk passively hits the outlet strut. The opening sounds thus yield direct information about outlet strut condition with minimal amount of disturbance caused by the energy radiated from the disk. Hence the opening sound is a very desirable acoustic signal to extract. Unfortunately, the opening sounds have much lower signal levels relative to the closing sounds and therefore noise plays a more significant role than during the closing event. Because of this it is necessary to screen the sounds for outliers in order to insure a high sensitivity of classification. Because of the sharp resonances appearing in the corresponding spectrum, a parametric processing approach is developed based on an autoregressive model which was selected to characterize the sounds emitted by the Bjork–Shiley convexo–concave (BSCC) valve during opening cycle. First the basic signals and the extraction process used to create an ensemble of heart valve sounds are briefly discussed. Next, a beat monitor capable of rejecting beats that fail to meet an acceptance criteria based on their spectral content is developed. Various approaches that have been utilized to enhance the screened data and produce a reliable heart valve spectrogram which displays the individual sounds (power) as a function of beat number and temporal frequency are discussed. Once estimated, the spectrogram and associated parameters are used to develop features supplied to the various classification schemes. Finally, future work aimed at even further signal enhancement and improved classifier performance is discussed.