|About GSBS | FAQ | Job Opportunities | Search UMDNJ|
Mohammad Khawar Zia
B.S., Rutgers University - 2006
Thesis Advisor: John L. Semmlow, Ph.D.
Graduate Program in Biomedical Engineering
Biomedical Engineering Building
599 Taylor Road, Room 126
Monday, August 27, 2012
Coronary artery disease (CAD) is the leading cause of death in the United States despite known ways to prevent and treat it. In CAD, the coronary arteries, which supply oxygenated blood to the heart tissue, gradually narrow and harden due to plaque deposition. This gradual narrowing and hardening can partially or completely occlude blood flow to the heart tissue. If the heart tissue is deprived of oxygenated blood for a long period, it is damaged or dies, leading to an event commonly termed a heart attack, which can lead to death. Progression of CAD can be controlled effectively using drugs and/or diet when the narrowing is not significant. However, in more than half the cases of CAD, no symptoms are present prior to a sudden heart attack that leads to death. Since symptoms of CAD are usually not present, CAD goes undetected and untreated, making it the leading cause of death. There is a clear need to develop a screening tool that can detect CAD in early stages when it can be treated effectively.
The acoustic approach to CAD detection offers a simple, cost-effective, and risk-free method for diagnosing CAD. This approach records heart sounds at the chest in an attempt to capture the faint acoustic signatures associated with turbulent flow through partially occluded coronary arteries. No other approach to detecting CAD can be as simple, cost-effective, and risk free, but the approach has yet to show sufficient accuracy for clinical use. A major limiting factor in the acoustic approach is its sensitivity to noise. Because the problem of detecting CAD sounds requires the resolution of a very faint signal buried in a substantial amount of noise, any additional noise, whether acoustic or electronic will reduce detection sensitivity.
In this thesis, a detailed analysis is performed on the signal and noise properties of a system designed to detect CAD sounds. The system was designed by SonoMedica, Inc., the industrial partner in this research. The primary goal of this analysis was to increase the signal-to-noise ratio (SNR) of the acoustic signal acquired by specially designed cardiac microphones placed on the chest. Accordingly, the first step in this project was to replace the original first stage instrumentation amplifier in the interface box with an amplifier that has lower combined voltage and current noise. The next step was to establish that the heart sounds captured at the chest using the improved data acquisition system were above the microphone and amplifier noise floor; i.e., that there was more signal than noise in the recordings (positive SNR).
An attempt was made to reduce the sensitivity of the microphones to external acoustic noise using two passive noise attenuation methods. The first method used a noise shield that covered the chest to block ambient noise. This noise shield reduced ambient noise by at least 5 dB below 400 Hz and by 20 dB above 400 Hz. However, the noise shield was inflexible and could not accommodate all subjects. As a result, its performance did not generalize to all subjects. The second method entailed the development of a rubber boot to enclose the microphone. This method gave modest noise attenuation, but it decreased the sensitivity of the microphone to high frequency signals. Both methods were found not suitable for clinical use and could not be used to mitigate acoustic noise.
This work also evaluated various data acquisition protocol parameters to find an optimal set of parameters that maximized the SNR of the recorded heart sounds. These parameters included patient position, strap tension for holding the microphones in place on the chest surface, impedance matching foam pad thickness, use of coupling gel, and breath hold. The SNR of the recorded heart sounds was determined by cross-correlating signals from two microphones placed side-by-side on the chest. The cross-correlation coefficient was converted to SNR using an empirically derived relationship. Parameters that gave higher SNR were selected for data acquisition.
Since the CAD sounds are maximal during diastole, this thesis also included the development of an ECG-based algorithm for isolating diastole that gives near perfect performance on non-ectopic heart beats. The excellent performance of the algorithm was, in part, the result of the algorithm exploiting a statistic computed over the entire ECG recording.
Finally, the most important contribution of this thesis is the development of a clinical noise detection algorithm that is robust to variation in background noise. Such an algorithm is crucial to the success of the acoustic approach. By automatically identifying and eliminating acoustic noise from heart sound recordings, the performance and ultimately the clinical utility of the acoustic approach to detecting CAD can be greatly increased. This algorithm can effectively detect and eliminate both external and internal noise from heart sound recordings. Depending on the configuration used, the algorithm showed that, on average, only 35% of a recording was noise-free. Using the ECG algorithm developed in this thesis to first isolate diastole, the noise detection algorithm showed that, depending on the configuration used, on average 35-70% of the diastole was clean in a recording. The methodology developed here will be essential to any system that uses heart sounds to diagnose coronary artery disease.