Abstracts

Rationale: While group-level differences in control structural connectomes and a variety of aggregated connectomes from patients with neurological disorders have been well investigated, individual variations in connectomes (which are crucial for neurosurgical clinical application while evaluating epilepsy patients) are understudied and the reference ranges within which network metrics for healthy controls vary are poorly quantified. This makes it difficult to use structural connectivity data derived from diffusion imaging to guide preoperative evaluation of epilepsy patients. 

Methods: The TractoFlow pipeline was used to conduct tractography on diffusion imaging data from one hundred unrelated subjects made available by the Human Connectome Project. The resulting tractograms (Figure 1) and a novel high-resolution brain atlas (Yale Brain Atlas) with 696 cortical brain parcels were used to generate structural connectivity matrices for healthy subjects, and the matrices were concatenated into an array (the reference structural connectome). A function was designed to take as input a patient’s structural connectivity matrix (derived using the same pipeline and atlas) as well as the reference structural connectome, perform matrix visualization, convert matrices to graphs, and conduct brain network analysis using commonly used network metrics at a whole brain level as well at a finer level (lobar/regional). Deviations in the structural connectome were captured using the Hamming distance between data from patient and controls. The reference range to capture normal individual variation was quantified by comparing individuals from the reference group to each other.

Results: We created a reference structural connectome that preserves individual level structural connectivity data using a set of healthy subjects, and designed a function that can use network analysis and distance metrics to extract deviations from the reference range values of healthy controls for a given individual patient connectome. The function also localizes these deviations to cortical parcels in a high-resolution anatomical brain atlas. Preliminary results captured the individual variation among the healthy control subjects’ structural connectomes and their network properties using a leave-one-out implementation of the function, where a single reference subject was compared to the rest of the cohort iteratively (Table 1). 

Conclusions: We have worked up the normal distribution and reference range of a few key brain network metrics for healthy subjects. When network metrics from our individual-data-preserving approach were compared to the standard aggregation approach, the network metrics for the averaged structural connectome differed significantly from all individual connectomes (supporting our claim that a methodology that preserves and quantifies this individual variation is beneficial for clinical application on patients). With the incorporation of patient data, we aim to apply this function to aid in the lateralization and localization of seizure onset areas, as well as understanding the structural connectivity within the seizure network of patients with medically intractable epilepsy. 

Funding: NIH award R01NS109062