Diffusion-Weighted Imaging (DWI) has demonstrated its potential in research applications for renal pathologies^(1,2). Compared to biopsy, DWI is non-invasive and the whole kidney can be imaged and assessed. However, renal multi-slice DWI acquisition has been considered too challenging because artifacts, such as motion, partly related to the multi-slice acquisition, limit the analysis^(3,4). Despite the use of physiological triggering schemes to limit respiratory artifacts, native kidney images can be impacted by the presence of signal dropout (local and non-consistent) causing slice variation. The mean of several slices or one single slice is commonly analyzed and variability between slices is lacking in the literature. We compared first the inter-slice variations between a standard fast gradient echo (GRE) and DWI sequences. Then, an algorithm inspired by a study minimizing bulk motion on cardiac DWI⁽⁵⁾ was proposed and tested to reduce the variability of Monoexponential Apparent Diffusion Coefficient (ADC) in renal multi-slice acquisition.

Four healthy volunteers were scanned on a PRISMA 3T MR (Siemens AG) with the spine and abdominal coils. The MR protocol included a GRE sequence with TR/TE=2000/95ms, resolution 1.4x1.4x5mm, GRAPPA acceleration factor 2, bandwidth 780Hz/pixel, 30 slices; and three single-shot SE-EPI DWI with sequence parameters TR/TE=2000/69ms, resolution 2x2x5mm, GRAPPA acceleration factor 3, bandwidth 2500Hz/pixel, 10 slices, respiratory navigator (PACE, prospective acquisition correction) and diffusion gradients (3 orthogonal directions, 10 b-values [0-900seconds/mm²]). Assessment of GRE and DWI slice variability before correction: One Region-Of-Interest (ROI) with both cortex and medulla was drawn on 4 consecutives slices of GRE and DWI. Care was taken to maintain similar locations over each sequence. In DWI, the ROI was drawn on b₀ images and propagated through all b-values with manual registration. The mean signal intensity (SI) of all voxels in the ROI was calculated. The Pearson standard deviation (PSD) between the SI of the 4 consecutive slices was used to quantify slice variation. To compare between sequences, GRE and DWI, PSD was expressed as the percentage of the mean SI. Motion correction algorithm: Registration of individual images and correction of signal dropout were performed with an in-house developed software using MATLAB®. The images of the 3 references acquisitions were registered for each b-values with gradient images. One acquisition was used as reference and the 2 others shifted pixel-by-pixel to maximize the gradient image product. From this, a reconstructed set of images was rebuilt with the pixel provided by the highest intensity value of the 3 acquisitions. Assessment of DWI slices variability after correction: PSD of the 3 reference acquisitions was averaged and compared to the PSD of the reconstructed images for each b-values. ADC variability: Quantitative ADC maps were also calculated on a voxel-by-voxel basis with OsiriX© ADC map plugin and used to compare the variability in ADC between slices before and after correction.Uncorrected, DWI images were impacted by signal dropout causing slices variations. This artifact was reduced in the corrected DWI compared to the 3 references DWI by the use of correction algorithm as shown in figure 1. The slice variations were on average 2.7% ± 2.9% in GRE images, 7% ± 1.2% [range: 2% to 17%] in uncorrected DWI and 4% ± 0.6% in corrected DWI. The signal dropout led to hyperintense area in ADC map as shown in the Multiplanar Reconstruction (MPR) of healthy volunteers kidney (Figure 2). Figure 3 represents the PSD of the uncorrected DWI (in red) with the error bars for the standard deviation between the 3 acquisitions, compared to the corrected DWI (in green) and the GRE (in blue) for each volunteer. In 36/40 b-values analyzed, the corrected DWI provided less variability than uncorrected with a reduced PSD between slices. The reduction of DWI slices variation led to a reduced slice variation in ADC map as shown in figure 4. We attempt to highlight the presence of signal dropout in DWI and ADC maps despite the use of a respiratory navigator. There is a real advantage in using a motion correction algorithm to reduce slice variations and improve the quality of renal DWI. An important factor gained is the reduction of signal dropout leading to less variability in multi-slice renal DWI. Despite the cost of 3 single-shot EPI acquisitions, this technique can reduce motion-induced artifact and provide robust renal DWI data for further quantification and detection of renal abnormality in clinical applications.