Validation of CT Images of Calcified Plaque
The ultimate goal of non-invasive plaque imaging is to characterize the distinct histological features of coronary artery plaque. In this regard, our research focuses on comparing CT images to histological cross-sections in order to rigorously assess the accuracy of CT-based identification and quantitation of calcified plaque in cadaveric hearts.
To register histology and CT, we use a combination of anatomical fiducial landmarks and physical CT parameters to identify cross sectional slices of exact matches. Using this technique we have successfully co-registered histology images to both 64-slice MDCT and high-resolution flat-panel volume CT. Due to the high x-ray density of typical calcified lesions, CT images of such lesions contain a characteristic “blooming” effect, yielding images of lesions which appear larger than their actual size. This effect appears to be more pronounced in the 64-slice MDCT than the fpVCT.
Cardiac CT Artifact Reduction
The non-invasive nature of cardiac CT angiography makes it ideally suited for screening purposes. The presence of calcification around the coronaries, however, makes it difficult to render some types of assessments, including stenosis and plaque characterization. Using novel and sophisticated image reconstruction techniques, we have preliminarily demonstrated the ability to limit the “blooming” effect of calcium.
This research endeavor involves several collaborating organizations, including MGH Cardiac MR CT PET Program and Center for Biomarkers in Imaging, Boston University, CenSSIS (Center for Subsurface Sensing and Imaging Systems), and CIMIT (Center for Integration of Medicine and Innovative Technology) as academic collaborators, along with several industrial partners.
Overcoming the Computational Barrier of Tomographic Image Reconstructions
While traditional filtered-backprojection (FBP) techniques for CT image reconstruction have served the radiological community well over the past several decades, the desire to push the limits of detection, coupled with the manufacturing of larger detector arrays, have increased the need for more precise image reconstruction techniques. Algebraic Reconstruction Technique (ART) and its numerous variants seek to overcome the approximations and the resulting artifacts of FBP. These techniques, however, are in general very computationally intensive.
We have undertaken multiple efforts to overcome the computational barrier. We have in place a state-of-the-art Beowulf cluster. We also utilize commercial high-throughput image processing systems. Most recently, we have collaborated with the Center for Sensing and Subsurface Imaging Systems (CenSSIS) at Northeastern University, and Boston University, to exploit graphics processing units (GPUs) to accelerate the ray-tracing computations associated with ART-based techniques.
|
 |
|
