O. Salvado, C. Hillenbrand, S. Zhang, J. Suri, and D. Wilson (USA)
MRI, inhomogeneity, atherosclerosis, classification, segmentation, FCM.
We are characterizing atherosclerotic disease in patients and animal models using multiple MR images having different contrasts such as T1-weighted, T2-weigted, and proton density weighted. We use intravascular and surface array coils giving good signal sensitivity but significant inhomogeneity. Because multiple images are obtained, it is desirable to acquire images quickly, limiting the signal to-noise ratio (SNR). In human carotid images, bias was corrected using a modified adaptive fuzzy c-mean method with a mechanical membrane model of the bias field [1]. Noise reduction filtering, background segmentation, outlier class identification, and signal normalization were all designed to address specific technical issues such as the noise, steepness of the sensitivity fall-off, and absence of signal near regions of interest due to fat suppression and blood flow compensation. In a synthetic image having a bias field measured from our MR system, variations across an area comparable to a carotid artery were reduced from 60% to <5% with processing while the misclassification rate was kept below 4% even with a poor SNR of <7. Human carotid images were qualitatively improved and large regions of skeletal muscle were quantitatively flattened and normalized for inter- and intra-subject variation. The method will facilitate interpretation of artery gray scales for manual plaque characterization and enable computerized plaque classification.
Important Links:
Go Back
