Frederick William Damen, Presenter: Nothing to Disclose

Qing Gao MS, Abstract Co-Author: Nothing to Disclose

Xiaohong Joe Zhou PhD, Abstract Co-Author: Nothing to Disclose

To compare the performance of a bi-exponential model and a statistical model (Yablonskiy, et al. (Magn. Reson. Med., vol. 50; 2003) for analyzing high b-value diffusion-weighted images of the human brain.

Diffusion-weighted images were acquired from healthy human volunteers on a 3 T MRI scanner using a customized echo planar imaging pulse sequence with an 8-channel head coil. Fifteen b-values ranging from 0 to 3,300 s/mm2 were produced by varying the Stejskal-Tanner diffusion gradient amplitude. At each b-value, the diffusion gradient was successively applied along each of the three orthogonal axes to acquire axial images. Trace-weighted images were then computed for analysis. The other acquisition parameters were: TR/TE=4000 /97.1ms, slice thickness = 5 mm, FOV = (22 cm)2, slices = 6, image matrix size = 1282, NEX = 6, and total scan time = 20 min. Maps of the diffusion parameters from each model were produced using customized software for all slices. Additionally, a region-of-interest (ROI) analysis was also performed on specific white matter and gray matter areas. The absolute percent error for each model was computed and used for performance evaluation.

At each voxel, the model that produced the smaller absolute percent error was retained and color coded. Analysis of the 6 contiguous slices showed that the bi-exponential model had lower errors in 79.8 ± 3.2% of the voxels (excluding the cerebrospinal fluid). In a slice containing the caudate nucleus and genu, the bi-exponential model had lower error for 78.7% of the brain parenchyma. In an ROI analysis over 3x3 voxels, the bi-exponential model produced percent errors of 3.2 ± 0.74% in the caudate nucleus and 5.9 ± 1.4% in the genu. For those same regions, the statistical model produced percent errors of 2.7 ± 0.49% and 8.2 ± 0.31%, respectively.

The bi-exponential model and the statistical model can both fit the high b-value diffusion data with good accuracy. The statistical model shows better performance in specific gray matter areas, while the bi-exponential model produces a better fit in white matter regions as well as some gray matter regions. Overall, more voxels of the brain parenchyma are better characterized by the bi-exponential model.

High b-value diffusion-weighted MRI with appropriate models can potentially provide valuable information on tissue structural changes during disease processes.

Damen, F, Gao, Q, Zhou, X, Comparison of Two Models for Analyzing Diffusion-weighted Brain Images with High b-Values.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8009813.html