Purpose: To develop a method to use information from multiple MRI contrasts to produce a composite angiogram with reduced sequence-specific artifacts and improved vessel depiction.
Methods: Bayesian posterior vessel probability was determined as a function of black blood (BB), contrast enhanced angiography (CE-MRA), and phase-contrast MRA (PC-MRA) intensities from training subjects (N = 4). To generate composite angiogram in evaluation subjects (N = 12), the voxel-wise vessel probabilities were weighted with a confidence measure and combined as a weighted product to yield angiogram intensity. For 23 internal carotid artery (ICA) segments (N = 23) from evaluation subjects, segmentation accuracy of composite MRA was evaluated and compared against CE-MRA using dice similarity coefficient (DSC).
Results: The composite MRA suppressed venous contaminations in CE-MRA, reduced flow artifacts, and velocity aliasing seen in PC-MRA and removed signal ambiguities in BB images. For ICA segmentations, the composite MRA improved segmentation over CE-MRA per DSC (0.908 ± 0.037 vs. 0.765 ± 0.079). Compared with CE-MRA, the composite MRA showed conservative changes in vessel appearance to small threshold changes. However, small vessels that are sensitive to registration errors or visible only weakly in CE-MRA were susceptible to poor depiction in composite MRA.
Conclusion: By dynamically weighting vessel information from multiple contrasts and extracting their complementary information, the composite MRA produces reduced sequence-specific artifacts and improved vessel contrast. It is a promising technique for semi-automatic segmentation of vessels that are hard to segment because of artifacts.
Keywords: Bayesian statistics; MR angiography; artifact suppression; multiple contrasts; vessel segmentation.
© 2019 International Society for Magnetic Resonance in Medicine.