Correction of eddy-current distortions in diffusion tensor images using the known directions and strengths of diffusion gradients

Diffusion tensor imaging (DTI) is an important tool for characterizing in vivo the anisotropic fiber structure within somatic tissues (1). The capacity of DTI to provide valid and reliable information about tissue structure, however, can be affected adversely by eddy-current artifacts. In echo-planar images, which are usually used to acquire diffusion tensor (DT) images, eddy currents produce significant distortion in the phase-encoding direction because of the relatively low bandwidth in that direction, and the large changes in diffusion gradients that occur during DTI. Image distortions from eddy currents blur the boundaries between gray- and white-matter tissues, lead to misregistration between individual diffusion-weighted (DW) images, and cause miscalculation of DTs. In general, eddy currents can be reduced effectively in one of three ways: 1) selecting the appropriate pulse sequences (such as a dual spin-echo sequence) (2,3) or gradient waveforms (such as bipolar gradients) (4), 2) correcting the k-space data (e.g., by calibrating eddy-current artifacts in k-space) (5-7), and 3) implementing postacquisition image processing to register the DW images to reference images. The third approach (implementing postprocessing algorithms) is appealing to most DTI investigators because of its relative ease and accessibility. One widely used postprocessing algorithm, iterative cross-correlation (ICC) (8), estimates distortions in DW images by cross-correlating them with an undistorted baseline image in terms of scaling, shear, and translation along the phase-encoding direction. The estimated distortion parameters are then used to correct the distorted images (8-11). One serious limitation of the original Haselgrove’s ICC algorithm (8), however, is its inability to correct image distortions at high b-values. The contrast of cerebrospinal fluid (CSF), gray matter, and white matter in images acquired with no diffusion weighting differs greatly from the contrast found in images acquired with high (b-value) diffusion weighting. This difference in contrast leads to unreliable registration of the two types of images, which in turn interferes with the correction of eddy-current distortions. Various methods have been proposed to remedy this problem. One method extrapolates distortion parameters from low to high b-value images (8). While other methods correct eddy-current distortions in high b-value images without extrapolating distortion parameters from low b-value images, these procedures undesirably require additional image acquisition (9-11) or extend scanning time, such as in the fluid-attenuated inversion recovery (FLAIR) sequence (12). This paper describes a new algorithm that detects eddy-current distortions by modeling those distortions with the known x, y, and z components of diffusion gradients. The method was validated in three experiments using a silicon oil phantom in which the contrast was uniform and relatively stable across a wide range of b-values. Finally, we applied this algorithm to correct distortions in images of the human brain and then assessed its effects on the resulting DT maps.