Abstracts

Rationale: Personalized network modelling of regional and inter-regional neural signalling has the potential to identify epileptic networks and plausible pathways of signal spread. Tractographic analysis of diffusion-weighted MRI has been the most common technique to infer structural connections in vivo, but these association tracts comprise less than 4% of cortico-cortical connectivity. Here, we constructed a novel model of cortical wiring that enriches diffusion MRI data with microstructural similarity (MS) and spatial proximity information. We tested whether our more comprehensive model of cortical connectivity would predict inter-regional coherence across multiple frequency ranges in epilepsy patients undergoing intracranial EEG implantations. Methods: Our approach was first developed in 100 healthy adults, then applied to four females with drug-resistant epilepsy. Participants underwent microstructural and diffusion MRI. We calculated three inter-regional connectivity matrices: MS from correlation of intracortical myelin profiles, tractography strength from spherical deconvolution informed tractography and spatial proximity from geodesic distance (GD). Matrices were concatenated and compressed into a new manifold space via non-linear dimensionality reduction. We calculated magnitude squared coherence between EEG contacts (at least 8 mins of rest), and mapped contacts into the connectivity-based subspace. Finally, we used adaboost regression to predict inter-contract coherence based on their distances in connectivity space. Results: In controls, the newly derived connectivity space depicts three axes of inter-regional differentiation, stretching out towards either sensory, prefrontal or limbic regions (Figure 1). The group wise pattern of structural wiring was consistent in single epileptic patients (Figure 2). Notably, distances in structural wiring between contacts explained up to 25% of variance in inter-contact coherence (R2=0.10±0.035). Across patients, the highest level of variance explained was in the higher gamma frequency band (80-250 Hz:). Using 70% of edges as training data and 30% of edges as test data, we found that structural wiring predicted coherence within approximately 1/3 of a standard deviation of real values (standardised mean squared error=0.36±0.21). Conclusions: Our multiscale model of structural connectivity provides a more holistic representation of cortical wiring, and is in closer conceptual contact to established models of cortical-cortical connectivity. Applying our framework to a epileptic patients undergoing intracranial EEG implantations, we could show promising utility in predicting inter-regional coherence measures at rest. Crucially, the model was found to be particularly powerful in measuring fast inter-regional phenomena, which may suggest utility in simulating ictal processes as well. Funding: Transforming Autism Care Consortium