To consider the role that axonal shape has on susceptibility-weighted imaging in white matter (WM), with application to demyelination.Recent work has described the magnetic susceptibility of myelin as a significant contributor to gradient echo signal.^1-6 Increasingly sophisticated biophysical models are employed to explain these signal properties, including effects like susceptibility anisotropy.⁵ This work investigates the effect of myelin shape on the modeling of demyelination in WM. First, a structural template of myelin is constructed from electron microscopy (EM) data of mouse WM. Next, simulations of the MR signal magnitude/phase are generated from this realistic template and compared to that of conventional models using circles. Finally, predictions of these two models are evaluated against experimental cuprizone mouse data, an animal model of demyelination.Magnetic field perturbations induced by densely packed axons with arbitrary geometries were modeled using the tensor formulation in the Fourier domain.⁷ These calculated field maps were used to simulate MR signal evolution in three compartments: extra-axonal, intra-axonal and myelin.^4,6 Relevant parameters for the compartments (proton density, T₂, magnetic susceptibilities) were based on literature values at 7T, Table 1.^5,6,7 Simulations considered both circular axons and more realistic geometries. For the latter, 2D EM data on mouse WM was acquired at 7.16nm resolution and hand-segmented. Packing properties between the EM and circle models are matched: both have a fiber density of 63%, the same Gamma distribution of axon sizes, myelin content and compartmental properties (Figure 1). Note that this rather low fiber density represents only myelinated axons. Demyelination is modeled for both geometries by gradually thinning the myelin structure, producing a range of g-ratios, from 0.71 (normal) to 0.98 (no myelin).Simulations were compared to experimental data acquire in a mouse model of demyelination. Nine mice were fed 0.2% cuprizone diet ad libitum for different durations ranging from 0-42 days to induce varying degrees of demyelination. Images were acquired ex vivo on a 7T animal scanner using a multiple echo gradient echo sequence with TEs=3,7,11...55ms (TR=1.5s, $$$70^\circ$$$flip, 80x80x300$$$\mu m$$$, 3 axial slices, 10 averages). Brains were scanned such that the corpus callosum (CC) was orthogonal to the static field, which is matched in simulation. Background field in phase images were removed using a ‘projection-onto-dipole-fields’ approach.⁸The simulated effect of demyelination on signal magnitude and phase is shown in Figure 3: circle model (top: A-B) and EM-based model (bottom: C-D). Each curve represents a different g-ratio, varying from 0.71 to 0.98. Both models predict amplified signal decay and phase evolution with increased myelination. At a given echo time, the circular model predicts slightly greater signal decay and much greater phase evolution than is predicted by the EM-based model. Given that these simulations were otherwise matched, their differences suggest that myelin geometry has a significant impact on both signal magnitude and phase across a range of demyelination stages.Demyelination was confirmed with Luxol fast blue histological stains (not shown). These stains are not sufficiently quantitative to establish whether duration on the diet was monotonically related to myelination, however there were clear differences in stain intensity between mice with long versus short-duration diets. Averaged signal magnitude and phase from an ROI in the CC (Figure 2) are plotted across the nine mice in Figure 4, color-coded by their cuprizone diet duration thereby presumed level of demyelination. There is a clear trend for faster signal decay and greater phase accumulation in mice undergone short diet durations (and therefore more myelin). At TE=55ms, the signal magnitude has attenuated to ~0.15–0.35 and the signal phase varies from ~-0.9-0 radians. These ranges are in good agreement with the predictions from the EM model (Figure 3), while signal predictions from the circular model are larger.Results in this work suggest that myelin geometry has a significant effect on the susceptibility-weighted signal. For the simulation parameters used here (based on literature), EM-driven geometries were more consistent with experimental data than circular geometries. While EM-based geometries are unlikely to represent a sufficiently generalizable framework for signal modeling, these results do have several important implications for interpreting these MR signals. First, literature estimates of these properties have generally been based on fitting a biophysical model to GRE-signals; our results suggest that assumption of circular geometries would bias susceptibility estimates. Conversely, even if the susceptibility values for myelin are known, circular models would led to consistent overestimations of myelin content since they produce larger magnitude and phase changes than realistic geometries. Sensitivity of the GRE-signal to axonal geometry therefore can be expected to bias estimates of microstructural properties from these signals.