Radiologists and physicists interested in correcting and assessing distortions on clinical images.Diffusion-weighted MRI (DW-MRI) employed to improve the detection of small cervical tumours suffers from distortion induced by susceptibility variation and severe eddy-currents. Eddy-current related distortions are caused by strong diffusion-sensitizing magnetic field gradients which have short ramp times and depend on the magnitude and direction of the diffusion encoding field gradients. This adds to the susceptibility induced distortions in EPI-based sequences where residual eddy-currents cause translation (B0 eddy-current field). To correct for spatial distortions on spin-echo MR images the forward and reversed gradient technique proposed by Chang and Fitzpatrick requires the acquisition of two images of the same object under the same conditions except for the polarity of the frequency- encode line. For DW-EPI, we have implemented a forward and reversed gradient technique in the phase-encode direction and investigate the clinical utility of the technique when viewed with T2-W images.Forty-one patients aged 25-72 years, (mean 40+11 years) with cervical cancer were imaged at 3.0T (Philips Achieva) using an endovaginal receiver coil developed in-house [2]. T2 Weighted (T2-W) TSE images were acquired in three planes orthogonal to the cervix (FOV 100mm, TE=80ms, TR=3400ms, SPIR fat suppression, left-right phase encoding, 2 averages, image acquisition matrix 288×288, 0.35mm resolution, 24 slices with 2mm slice thickness and 0.2mm separation). Diffusion-weighted images were acquired coronal to the cervix (single shot SE EPI, FOV 100mm, TE=52ms, TR=8000ms, SPIR fat suppression, left-right phase encoding, 1 average, band-width 9.6Hz/pixel, EPI factor 115, acquisition matrix 80×80, 1.25mm in-plane resolution, b-values 0, 100, 300, 500 and 800s/mm2, 24 slices with 2mm slice thickness and 0.2mm separation). A left-right phase encoding was chosen rather than anterior-posterior encoding to avoid artefacts through the cervix and the gradient reversal method applied along the phase-encode (left-right) axis. Acquisition time for the diffusion-weighted images was 4 min and 33 sec for each gradient direction. Corrected images were generated using the Chang-Fitzpatrick algorithm in MATLAB (The MathWorks, Natick, Massachusetts). Images were scored as positive, negative or equivocal for tumor by two independent observers of 20 years and 3 years experience of endovaginal MRI respectively on viewing the T2-W and left gradient images together and the T2W and right gradient images together. In the equivocal group, the T2-W+corrected images were viewed together and scored either as positive or negative for tumor. All cases that were scored as positive for tumor by the experienced observer had tumor volume measured on the T2-W images by drawing a region of interest around an intermediate signal intensity lesion that showed corresponding diffusion restriction and multiplying the sum of the areas by the slice thickness.Tumor volume in these patients ranged from 0.07-18.4cm³ (mean+SD 4.4+5.6cm³). Classification for tumor presence on uncorrected and corrected images are given in Table 1 along with intraobserver agreement. Sensitivity and specificity against subsequent histology for corrected images was 100% (8/8) and 50% (3/6) for observer 1 (Figure 1) and 50% and 20% respectively for observer 2. Tumor volume for the uncorrected images classified as positive was 4.4+5.6cm³, for the images classified as equivocal it was 0.18+0.44cm³.Phase-based methods of distortion correction [3] yield good corrections but they suffer from increased acquisition time. For our work, the method of Chang and Fitzpatrick offered the simplest solution, albeit requiring an extra image acquisition for the patient. It also had the advantage that the corrected images generated from the combined distorted left and right gradient images were expected to exhibit a signal-to-noise ratio higher than either of the distorted images [4]. Also, despite the limitation that an extra sequence was required incurring a time penalty, the addition of distortion correction enabled the 34% of equivocal cases (maximum tumor volume 1.5cm³) to be further classified with high sensitivity. The low specificity is likely related to the limits of the spatial resolution of the technique in relation to tumor volume. The improvement in diagnostic performance of the corrected images was primarily of benefit to an experienced compared to a less experienced observer. Distortions encountered with an EPI based diffusion-weighted sequence were particularly pronounced in our study because of the B0 field inhomogeneity induced by the endovaginal coil, but the correction methods described here may be applied to any situation or commercial coil where excessive B0 distortion is encountered. Application of the reverse gradient algorithm for distortion correction of echo-planar diffusion-weighted endovaginal images improved diagnostic performance in a third of cases for a more experienced observer.