Abstracts

Rationale: As epilepsy is progressively understood as a network disorder, connectome analyses are increasingly used in epilepsy research and facilitate our understanding of the brain as a network. Diffusion Tensor Imaging (DTI) and tractography allow the visualization and analysis of the brain’s structural connectome. However, DTI and tractography data processing is complex, with multiple options at each step that may influence final results. One of the major options is the choice of tractography algorithm used to reconstruct white matter tracts. Currently, there is very limited data on the impact of tractography algorithms on subsequent connectome analysis. In this study we compare two representative tractography algorithms and evaluate their impact on subsequent connectome measures.

Methods: DTI data from a 3T Scanner with 64 diffusion weighted directions was analyzed from 14 healthy controls and one epilepsy patient with a left temporo-occipital malformation. 226 anatomical regions (210 cortical, and 16 white matter) were defined by the Brainnetome and Desikan-Killiany atlas and were coregistered to every individual's T1 scan and expanded 3mm into the white matter. These regions are the nodes of the connectome, with the number of fibers between regions as the weighted edges. Two different whole brain tractography algorithms were used: one deterministic (Tensor_Det, TD) and one probabilistic (two-tensor using an Unscented Kalman Filter, UKF). Connectome analyses were done with the Python implementation of the Brain Connectivity Toolbox. We compared two measures of connectivity: global and local efficiency, as these are frequently used in epilepsy research.

Results: Global efficiency values differed significantly between the two algorithms within the control group. With TD tractography, global efficiency values ranged from 3.48-4.34, with a mean of 4.02 (SD=0.32), with UKF from 1.95 to 3.14, with a mean 2.43 (SD = 0.3). This difference was statistically highly significant (p < .0001). Global efficiency in the patient was 3.94 with TD and 1.38 with UKF, only falling out of normal range with UKF tractography.  Local efficiencies also differed significantly between the two algorithms: the ratios of UKF/TD local efficiency ranged from as low as 0.02 up to 1.14, with a mean of 0.32. Some regions, such as the inferior temporal gyrus showed larger differences (ratio 0.21) than others (precentral gyrus: ratio 0.48). Local connectivity measures in the patient also differed between the algorithms: with UKF tractography, the left parahippocampal gyrus showed a 4-7 fold increase, but not with TD.