We present an innovative paradigm to overcome artifacts of individual MR angiography techniques by utilizing complimentary information existing across multi-contrast MR images. This technique applies Bayesian statistics to extract vessel likelihoods from each image type and generates a single ‘composite’ angiogram. Composite angiograms are computed utilizing black blood (BB), contrast enhanced MRA (CE-MRA), and phase contrast MRA (PC-MRA) images acquired in subjects with known neurovascular disease. The composite angiogram is demonstrated to improve vessel lumen depiction overcoming artifacts in individual source images from background enhancement, air cavities, and flow in CE-MRA, BB, and PC-MRA, respectively.
The proposed method determines the probability that a particular voxel is a vessel ($$$v$$$) given its intensity values in BB ($$$X_{\text{BB}}$$$), CE-MRA ($$$X_{\text{CE}}$$$), and PC-MRA ($$$X_{\text{PC}}$$$) images, namely $$$P(v|X_{\text{BB}}X_{\text{CE}}X_{\text{PC}})$$$. This probability is not readily available but Bayesian statistics allows its calculation in terms of more available probabilities:
$$P(v|X_{\text{BB}}X_{\text{CE}}X_{\text{PC}})=P(v)\frac{l(v|X_{\text{BB}})l(v|X_{\text{CE}})l(v|X_{\text{PC}})}{P(X_{\text{BB}})P(X_{\text{CE}})P(X_{\text{PC}})}$$
where P(v) is the prior probability of vessels, and $$$l(v|X)$$$’s are the vessel likelihoods given intensity X in the corresponding image. Unlike $$$P(v|X_{\text{BB}}X_{\text{CE}}X_{\text{PC}})$$$, $$$l(v|X)$$$'s can be easily estimated with known locations of vessels in training data sets, obtained either through manual or unbiased automatic segmentation. The final vessel probability reflects vessel likelihoods given by each MR image considered, and yields improved angiogram which fully extracts complimentary information among them.
High resolution BB, CE-MRA, and 4D-flow intracranial scans were performed on 7 human subjects with known neurovascular disease, using a 3T scanner (MR750, GE Healthcare, WI, USA) with a 32-channel head coil (Nova Medical, MA, USA). 3D BB were collected utilizing a DANTE1 prepared spin echo sequence with 0.75x0.8x0.8mm3 resolution, while CE-MRA and 4D flow MRI were acquired with center out 3D radial sequences2-3 with 0.6mm isotropic spatial resolution. BB and CE-MRA images were first registered to the 4D-flow magnitude images using 3D rigid registration with a mutual information (MI) metric4 (ANTs5). BB, CE-MRA, and PC-MRA images of all subjects were scaled to have the same intensity scales by matching cumulative distribution functions (CDF) for each image type.
In one subject, likelihood statistics were compared between exhaustive manual segmentation and semi-automatic sampling of vessels based on CE-MRA and PC-MRA images. Manual segmentation was accomplished with custom multi-contrast segmentation tool developed in MATLAB (Mathworks, Natick, MA, USA). For the semi-automatic sampling, vessels were sampled based on histogram thresholding. PC-MRA images were used to yield CE-MRA likelihoods, CE-MRA images to yield PC-MRA likelihoods, and CE-MRA and PC-MRA to yield BB likelihoods. This assumes the artifacts in the images to be independent from one another.
Likelihood statistics were then obtained and averaged from 4 subjects comprising the training set using automatically sampled vessels. Composite angiograms were obtained for the training-group subjects by computing vessel probabilities as well as the remaining 3 non-training datasets. Images were interrogated for image artifacts and qualitatively assessed for overall image quality.
1. Xie Y, Yang Q, Xie G, Pang J, Fan Z, Li D. Magn Reson Med. 2016;75(6):2286-2294.
2. Johnson KM, Lum DP, Turski PA, Block WF, Mistretta CA, Wieben O. Improved 3D phase contrast MRI with off-resonance corrected dual echo VIPR. Magn Reson Med. 2008;60(6):1329-1336.
3. Johnson KM, Fain SB, Schiebler ML, Nagle S. Optimized 3D Ultrashort Echo Time Pulmonary MRI. Magn Reson Med. 2013;70(5):1241–1250.
4. Kim D, Hoffman C, Wieben O, Johnson KM. Flow and structure with simultaneous visualization of registered 4D flow and black blood MRI. ISMRM 24th Annual Meeting & Exhibition 2006; 2701.
5. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC. A reproducible evaluation of ANTs similarity metric performance in brain image registration. Neuroimage. 2011;54(3)2033-2044