The importance of atherosclerotic disease in coronary artery has been a subject of study for many researchers in the past decade. In brief, the aim is to understand progression of such a disease, detect plaques at risks (vulnerable plaques), and treat them selectively to prevent mortality and immobility. Consequently, several imaging modalities have been developed and among them intravascular ultrasound (IVUS) has been of particular interest since it provides useful information about tissues microstructures and images with sufficient penetration as well as resolution. In general, the ultimate goal is to provide interventional cardiologists with reliable clinical tools so they can identify vulnerable plaques, make decisions confidently, choose the most appropriate drugs or implant devices (i.e. stent), and stabilize them during catheterization procedures with minimal risk. In this work, we review existing atherosclerotic tissue characterization algorithms including the state-of-the-art virtual histology (VH) framework, which has been implemented in the Volcano (Rancho Cordova, CA) IVUS clinical scanners using 64-elements 20 MHz phased-array transducer. Initially, we intended to extend this technique for data acquired with 40 MHz single-element transducers. For this reason, we started acquiring in vitro IVUS data and studied involved challenges from specimen preparation toward classification. We observed inconsistency among extracted features along with transducer's spectral parameters (i.e. bandwidth, center frequency). This, in addition to infeasibility of construction of reliable training dataset due to heterogeneity of atherosclerotic tissues motivated us to develop an unsupervised texture-based atherosclerotic tissue characterization algorithm. We proposed a two-dimensional multiscale wavelet-based algorithm to expand IVUS backscattered signals and/or grayscale images onto orthogonal symmetric quadrature mirror filters (QMF) such as Lemarie-Battle. At the bottom of decomposition tree, we employed ISODATA to cluster enveloped detected features in an unsupervised fashion and classify atherosclerotic plaque constitutes into fibrotic, lipidic, calcified, and no tissues. For the first time, we studied numbers of factors that were necessary for extension of in vitro derived classifier for in vivo applications such as reliability of classified tissues behind arc of calcified plaques and effects of pressure changes as well as flowing blood on constructed tissue color maps, called prognosis histology (PH) images. The second half of this dissertation is devoted to automatic detection of lumen borders in IVUS grayscale images acquired with high frequency (40 MHz up) transducers where more scattering exhibited within lumen area that makes the problem of interest more challenging. We established our framework on three-dimensional expansion of IVUS sub-volumes onto orthonormal brushlet basis function. The rational behind our framework was presence of incoherent (i.e. blood) versus coherent (i.e. plaque, surrounding fat) textural patterns along pullback direction, which was motivated by what an interventional cardiologist does to locate the lumen border visually by going back and forth among IVUS frames. We studied the feasibility of brushlet analysis through filtering blood speckles and supervised classification of blood versus non-blood regions. Our preliminary study confirmed that the most informative features reside in the innermost cubes, representing low-frequency components in transformed domain. Finally, we explored that tissue responses to IVUS signals are proportionally preserved in brushlet coefficients and it enabled us to classify blood regions in complex brushlet space. Subsequently, we employed surface function actives (SFA) to estimate the lumen borders after regularization. In a comparison study, we quantified our results with two of existing algorithms, employing IVUS grayscale images acquired with 40 MHz and 45 MHz single-element transducers. Overall, our proposed algorithm outperformed and the resulting automated detected borders showed good correlation with manually traced borders by an expert.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8FR03S6 |
| Date | January 2011 |
| Creators | Katouzian, Amin |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0024 seconds