An information-based comparison of diffusion attenuation models in normal and inflamed bone marrow.

Diffusion-weighted imaging has received attention as a method for characterizing inflammatory exudates in bone marrow in immune-mediated inflammatory diseases and reveals an increase in diffusivity in regions of bone marrow oedema. Various models of diffusion attenuation have been investigated but the model providing the best description of tissue pathophysiology in regions of marrow oedema is unknown. Determining the most appropriate model is an important step towards protocol optimization and the development of a robust and clinically useful method. We aimed to determine which of three candidate models of diffusion attenuation most accurately describes the acquired signal from normal and inflamed bone marrow. 11 subjects with spondyloarthritis and evidence of active inflammation (ie bone marrow oedema) on MRI and 17 patients with no evidence of active inflammation underwent diffusion-weighted imaging of the sacroiliac joints (b-values 0, 50, 100, 300 and 600 s/mm ). Monoexponential, intravoxel incoherent motion (IVIM) and kurtosis models were fitted to the acquired signal from regions of interest in areas of bone marrow oedema and normal marrow. The three models were compared in terms of sum of squared error and information content (corrected Akaike information criterion). Model parameters were compared between regions of bone marrow oedema and regions of normal marrow. f the three models investigated, the IVIM model provided the best description of the signal over the 0-600 s/mm range across normal and inflamed bone marrow. There was a particular advantage of the IVIM model in normal marrow, where it was best able to capture the pronounced fast diffusion component observed in several cases. However, IVIM and kurtosis effects both became smaller and the signal behaviour became closer to monoexponential in the presence of bone marrow oedema. Our data suggest that increases in D (in the IVIM framework) might account for the reduced deviation from monoexponential behaviour in oedematous bone.