The virtual multiple sclerosis patient: on the clinical-radiological paradox

Abstract

AbstractMultiple sclerosis (MS) is typically diagnosed based on the clinical presentation, the presence of structural MRI lesions, and a “no better explanation” criterion. The structural lesions, disseminated in time and space, are a consequence of autoimmune processes leading to the damage of the myelin sheath in the central nervous system. As such, one would expect that more lesions would relate to higher clinical disability. However, a conflicting scenario is often present, with a high lesion load related to mild clinical impairment, and vice versa, a phenomenon referred to as the “clinico-radiological paradox”. The myelin damage in MS is widespread, which is likely mirrored in a widespread slowing of conduction velocities. However, conduction velocities are typically measured on selected white-matter tracts (e.g., visual evoked potentials), which do not directly relate to clinical impairment. In this paper, we hypothesize that the overall slowing of conduction velocities (i.e., across all brain tracts) is a better predictor of clinical disability. However, estimating the whole-brain average velocities is challenging. To overcome this obstacle, we estimated patient-specific conduction velocities in MS patients by merging multimodal data (i.e., DTI and source-reconstructed magnetoencephalography) to inform large-scale brain models, fitted on each individual patient. We started from the known reduction of the power of the alpha frequency band, as well as the shift in its peak, observed in MS patients. We then reproduced these individual spectral features in silico using large-scale models based on the individual connectomes. We then used state-of-the-art deep neural networks for Bayesian model inversion to estimate the most likely average conduction velocity in each patient, given the observed spectral features (and the connectomes). Finally, we used the inferred conduction velocities to predict the individual clinical disability. We find that the conduction velocities inferred for patients are significantly lower than those inferred for controls and that they are predictive of individual clinical disability, well above the predictive power of demographic and clinical variables and lesion load. Our results suggest a biologically and physically plausible solution to the “clinico-radiological” paradox, where the inferred, individual changes in conduction velocities across the whole networks are proposed as causative to the clinical disability.