Diffusion weighted (DW) MRI is increasingly used in diagnostics.¹ Echo-planar imaging (EPI) is often employed to minimize scan time, but variations in the magnetic field (B0) geometrically distort EPI images due to a low pixel band width in the phase-encoding (BW_PE) direction. More importantly, information is lost, when signals from distinct, possibly non-neighboring, voxel locations are reconstructed into the same voxel resulting in signal pile-up. Geometric distortions can be corrected using a measured B0 map or by combining EPI images obtained with opposite gradients.² However, these methods do not correct for signal pile-up. Our objective is to fully correct DW-EPI images using a combination of the above methods.

On a 3T MRI, we acquired two EPI series with opposed PE gradients in RL direction and a 3D dual spoiled gradient echo sequence (SPGR) to map B0. Both EPI images were corrected for geometric distortions by the standard correction method derived from B0 and BW_PE.³ Using this information, we created distortion maps, which show how the EPI signals are distributed in PE direction by B0 variations (figure 1).⁴ Signal pile-up occurred when distinct locations or several neighboring voxels were mapped onto the same distorted voxel. The distortion-corrected images were averaged into a single image rejecting voxels with signal pile-up. These voxels contain data from only one EPI image.

First, the correction method was demonstrated on an ice-water phantom consisting of six tubes with sucrose solution around an air cavity. The fully corrected images were compared to the raw EPI images and the images corrected with the standard method. The maximum signal intensity in ROIs inside the tubes originated from the voxel with largest signal pile-up, whereas the undistorted signal was approximated by the median value. Additionally, the change in signal homogeneity was determined by the coefficient of variation (CoV defined as standard deviation divided by signal mean). ADC values were determined from the raw and fully corrected EPI images.

We applied the correction method in two patients with prostate and rectal cancer and visually inspected the raw and corrected images. In regions with clear signal pile-up, the maximum signal intensity was compared to the maximum value in the fully corrected image.

For the EPI sequences, BW_PE differed from 5.9 Hz/mm in the phantom measurement (SENSE=1), to respectively 9.6 Hz/mm and 13.3 Hz/mm in the rectum and prostate protocols (SENSE=2). The phantom was scanned with b-values of 0 and 200 s/mm² to determine ADC maps. For the patient data, the SPGR sequences were adjusted to remain in reasonable scan times, resulting in different voxel sizes and FOV.

The SPGR data of the phantom are presented in figure 2. Figure 3 shows the raw and corrected EPI images. Most distortions and intensity variations within the tubes were removed using the standard correction method. However, in regions with signal pile-up this method did not suffice. This is shown by signal variations within the tubes, for example in the indicated ROI. Here, the maximum signal intensity reduced from 7.4 10⁵ in the raw image to 4.9 10⁵ in the B0 corrected image and 3.9 10⁵ in the fully corrected one, accompanied by only a minor change in median values from respectively 3.1 10⁵ to 2.7 10⁵ and 2.8 10⁵. The increase in homogeneity was quantified by the change in CoV in this ROI. While the CoV was similar for the raw EPI image and B0 corrected image (0.34 and 0.35), it decreased to 0.12 after applying the full correction. The ADC maps in figure 4 demonstrate the retrieval of the geometry after correction. Here, the effect from signal pile-up is limited, since most piled-up intensities originated from locations within the tubes.

In the patient data (figure 5), an apparent intensity asymmetry in RL direction around the rectum was seen in the raw EPI images. Its sign changed in the image with opposed gradient, indicating its origin was signal pile-up. After applying the full correction, signal pile-up was removed. This is illustrated by the change of the maximal signal intensity in the region of the rectum. Here, the maximum was lowered from 1.0 10⁴ to 6.4 10³ and from 6.1 10³ to 3.7 10³ for the rectal and prostate cancer cases.

We demonstrated a method to fully correct EPI images. It does not only correct for geometric distortions, but also for possible signal pile-up near gas pockets, which obscures the real diffusion signal. Corrected DW-EPI images can improve the identification of healthy and abnormal tissue near these regions.