In this chapter, we discussed some postprocessing techniques to provide reliable and practical solutions for carotid plaque study based on MR images. For the three categories of images, single contrast weighting gray level images, image sequences, and multiple contrast weighting images, we have developed algorithms to address their specific needs and integrated into a software package, quantitative vascular analysis system (QVAS).
For single contrast weighting gray level image, we use MAP criterion with MRF priors as a powerful tool to build up the image model. The inherent noise-resistant ability and explicit description of pixel relations guarantee the results reliability and robustness. Also, the QHCF algorithm provides a feasible solution for its implementation in practical applications.
The solutions for image sequence segmentation and object tracking are built on the MRF-based active contour model. This framework incorporates the accurate and reliable region segmentation of MRF with the optimal contour tracking ability ofminimalpath approach. To ensure the optimal combination ofthesetwo models, a new criterion, maximum reliability, is set up as a bridge. This framework is also very flexible and extensible to include additional prior knowledge for various applications. In this study, it has been successfully applied to carotid artery tracking and lumen segmentation in MR image sequences.
Our initial study on multiple contrasts weighting MR image segmentation extends the MRF to multiple dimensions. However, because of the intrinsic limitations of this model, we adopted and further enhanced a clustering-based algorithm by employing mean shift as density estimator. The results of multiple contrast weighting MR image segmentation and the histology section validation demonstrate very successful performance.
Detection of FC status is crucial for understanding the disease status and prognosis of atherosclerosis. The preliminary algorithm introduced in section 8.5 shows promise in separating stable (thick) and unstable (thin) FCs. Future work is aimed at improving the detection of ruptured cap and differentiating it from thin caps.
Since the images in our study are of poor quality than are usual practical images, the algorithms for gray level images and image sequences segmentation can be applied as general solutions. The multiple contrast weighting approaches can also be used for color images segmentation because their general properties are shared.
Was this article helpful?