Passive Auscultation Noninvasive acoustic characterization of thoracic pathologiesAuscultation sounds contain vast amounts of diagnostically useful information that has been largely untapped since the invention of some of the advanced medical imaging modalities like echocardiography, ultrasound imaging, CT, MRI, etc. A simple array of acoustic electronic sensors mounted on the human torso can form an excellent low-cost, noninvasive, light weight, portable alternative diagnostic system that could find applications even in out-patient home-monitoring systems and battlefield operations. However, there are a few challenges that need to be addressed before such a system could be put to practice.We require signal processing / pattern recognition algorithms that can characterize the sound measurements to provide an accurate diagnosis. Recent times have seen efforts focused towards this task with little or no success either due to the lack of understanding of the actual physics of sound propagation in the thoracic cavity or because of the use of overly simplistic models for signal processing. In order to gather a better understanding of the propagation of intra-thoracic sounds to the chest, a finite-difference time-domain (FDTD) forward model was developed that simulates the underlying physics by solving the viscoelastic wave propagation equations for the thoracic geometry. The model can thus provide simulated auscultation signals corresponding to various intra-thoracic sounds and thoracic pathologies that can subsequently be used for development of better signal processing / pattern recognition algorithms. A simulation of the propagation of a gaussian first derivative forcing pulse (center frequency 100 Hz) from the mitral valve location in the human thoracic cavity is provided here. The spatio-temporal distributions of the vertical component of the velocity (Vz), magnitude of velocity (Vmag), vertical normal stress (log |Tzz|) and shear stress (log |Txz|) are shown as the wave propagates from the mitral valve towards the surface of the thorax. Click on the variable name to view the corresponding video.
The array of sensors signals can be used to: (a) localize the intrathoracic sounds - determine the origin of the sound sources
Click here to view the poster that was presented at this symposium. The idea was also shared amongst other graduate students at the Graduate Academic Conference (GAC 2009) sponsored by the College of Graduate Students (COGS) at Michigan State University. Noninvasive detection of fractures in prosthetic heart valvesThe Bjork Shiley Convexo-Concave (BSCC) metallic prosthetic heart valve was one of the most popularly implanted valves in patients diagnosed with stenosis or incompetence between 1979 to 1986. The valve contained two metallic struts that held the occluder disk (made of pyrolitic carbon) in place. While the inlet strut was integral to the suture ring of the valve, the outlet strut was welded to the structure. Finite element modeling studies of the BSCC valve revealed that the acoustic resonant modes of an intact valve (~7kHz - 9kHz) distinctly differed from that of a SLS valve (~2kHz - 4kHz). Thus, determining the resonant frequency was an indirect way to diagnose the condition of the valve. Subsequently, two noninvasive methods to determine the resonant frequency of the valve were developed.
As the name suggests, the method employed auscultation sounds to determine the condition of the valve. A pulse duplicator system was used to simulate the valve beat sounds, and an electronic stethoscope was used to record the signals. The sound signals were analyzed in the spectral domain. In particular, a short-time Fourier Transform was used to extract energy based features in the two different frequency ranges, 2kHz - 4kHz and 7khz - 9kHz. Thresholds were determined based on experimental studies on 34 (19 intact, 14 SLS) valves to differentiate between the two classes using a clustering algorithm.
The EMAT technique is an elegant method to acoustically excite the valve with a desired force intensity, at a chosen frequency using low-intensity electromagnetic fields. The method uses two orthogonally aligned electromagnets, operating at frequencies f1 and f2 (say). While one of the electromagnetic fields induces a current in the suture ring of the valve, the other interacts with this current and generates a Loretz force at the beat frequency |f2- f1|. Thus, if f1 is fixed at say 30kHz, and f2 is swept from from 31kHz to 40kHz the valve can be made to vibrate from 1kHz to 10kHz. By measuring the displacement or velocity vibrations generated by the valve in this frequency range, one can determine the resonant frequency, and hence the condition of the valve. A laser doppler vibrometer was used to measure the vibrations of the valve. Automated analysis of eddy current data for nondestructive assessment of steam generator tubes
Noninvasive diagnosis of Gastro-Esophageal Reflux Disease (GERD) in preterm infants
Noninvasive assessment of structural integrity of SOFI foam used in space shuttles
Forward and Inverse Problem of ElectrocardiographyUnderstanding the biological causes of cardiac arrhythmias remains a challenge to the medical community to this day. A noninvasive imaging technique, termed as electrocardiographic imaging (ECGI) that attempts to reconstruct the epicardial potential distributions through measurements of body surface potentials has thus been a potential area of interest. This problem of imaging (an inverse problem) however is highly ill posed. The project attempts to address this inverse problem of ECGI, firstly by designing a forward problem model, and then solving the inverse problem through various regularization techniques. Two of these, namely the Generalized Cross Validation (GCV) technique and the Composite Residual and Smoothing Operator (CRESO) technique have been discussed in detail and tested out on simulated data. Results show considerable promise in these two techniques. Click here to view the project report. Automated analysis of ECG waveforms for detection of Ventricular Arrhythmia
|
