Chemical shift fat fraction measurement in muscular dystrophy provides an objective non-invasive method of assessing the efficacy of gene therapy trials, independent of patient motivation¹. Duchenne muscular dystrophy is a devastating genetic disease which manifests in boys from the age of 3-5 and progressively destroys skeletal, cardiac and respiratory muscles. It causes the replacement of muscle by fat and fibrosis, and is now the subject of first-in-human therapy trials. The varying motivation of the boys makes traditional physical function tests too variable for sensitive clinical trial evaluation, so objective MRI measures of muscle involvement have become important. However, the boys have limited tolerance for keeping still in the scanner and the cost of lengthy MR acquisitions limits their use. In this study, we validate highly accelerated fat fraction measurements using data undersampling and combined compressed sensing and parallel imaging (CS-PI) reconstruction in adults with the similar, but slower progressing disease, Becker muscular dystrophy. We looked in particular at the optimization of reconstruction parameters across an adult cohort, the image quality of accelerated images and the fidelity of the muscle fat fractions.

Fully-sampled and undersampled data were acquired on a 3T Philips Achieva and a 6 channel cardiac coil using a custom 3D gradient echo with variable density Poisson disk undersampling at 3.65´, 4.94´ and 6.42´ accelerations ². A 256´190´48 matrix (1.25´1.0´5mm) was used with TR/TE/FA = 10ms/4.40,5.18,5.96ms/3^o. The acquisition times were 273s (full) and 75s, 55s and 43s (undersampled). 8 patients with Becker muscular dystrophy were recruited under local ethical approval and informed consent. Imaging blocks were acquired for calf and thigh muscles. Undersampled data were reconstructed with CS-PI using an L1-ESPIRiT approach ³, optimizing the reconstruction parameters as previously described ⁴. A coherent 3.36x undersampling was acquired for GRAPPA (PI) reconstruction alone. Fat fraction maps were calculated with a 6-component fat model and single component R2* modelling ⁵. Regions of interest were drawn on 10 axial sections on the fully sampled left leg for 12 anatomical areas (Table 1). The mean fat fraction was calculated for each anatomical area on the fully and undersampled reconstructions. 940 pairs of data were successfully evaluated to derive Bland-Altman parameters of bias and 95% limits of agreement. To assess image quality, two independent observers evaluated the percentage of anatomical ROIs that could be successfully delineated on each accelerated reconstruction.

Figure 1 shows that excellent image quality was obtained with CS-PI reconstruction. The optimal reconstruction parameters were found to be consistent between acceleration factors and across the patient cohort ³, providing confidence to use the accelerated protocol prospectively without fully sampled data. Table 1 shows the fat fractions were faithfully reproduced in the accelerated scans and the Bland-Altman analysis of fat fractions (Table 2, left) showed no net bias and demonstrated that CS-PI reconstruction outperforms PI alone. The two observers were able to confidently identify all ROIs in the 3.65x and 4.94x CS-PI images, but not in the 6.42x CS-PI image, where subtle features were absent, or the 3.36x PI only image (Table 2, right).