Abstracts

Rationale: In patients with refractory temporal lobe epilepsy (TLE) surgical interventions such as anterior temporal lobe resections (ATLR) and laser interstitial thermal therapy (LiTT) can be an effective treatment. In these interventions, select white matter (WM) fibre bundles are at risk of damage with resultant functional deficits, such as a visual field deficit by damaging the optic radiation. Fibre tractography during surgical planning can help prevent damaging these fibres but has significant inter-rater variability. We present an automated method for fibre tractography of bundles at risk in TLE surgery to ensure more consistent tracts. Methods: 10 TLE patients planned for ATLR (mean age 42.5y; 4F) underwent imaging on a 3T GE MR750 including a 3D-T1 and a multi-shell diffusion MRI protocol with 2mm isotropic resolution and 11, 8, 32, and 64 gradient directions at b-values of 0, 300, 700, and 2500 s/mm², respectively. T1-weighted scans were segmented into 208 regions using Geodesic Information Flows (GIF) into 208 regions. Fibre tractography was performed based on WM fibre orientation distributions (FODs) from multi-shell multi-tissue constrained spherical deconvolution (CSD) in MRtrix3. For the optic radiation (OR), tractography was performed from an LGN ROI, with our automated method using Geodesic Shape-Based Averaging to create the ROI from manual delineations on 59 previous datasets, including only those fibre tracts ending in the occipital cortex. Arcuate, inferior longitudinal, inferior fronto-occipital, and uncinate fasciculi (AF, ILF, IFOF, and UF, respectively) were obtained by tractography between their cortical termination areas. Two manual raters (M1, M2) reconstructed these five fibre bundles to compare to our automated approach (Au). Tractography maps were generated, thresholded to remove the lowest 2%, and binarised. For all bundles, spatial agreement between methods (M1-Au, M2-Au, M1-M2) was quantified by Cohen’s kappa, and differences between the three groups compared using ANOVA.For the OR, localisation of Meyer’s loop (ML) is the most clinically relevant information, so we evaluated the distance from temporal pole (TP) to Meyer’s loop: TP-ML. This was calculated as a relative value by subtracting the shortest TP-ML of any method (M1, M2, Au) from the TP-ML, for each subject and side. Results: Spatial agreement is summarised in Table 1, showing a significantly higher agreement between M1 and Au than for M2-Au (p=8.6e-7) or M1-M2 (p=9.2e-7). Example tracts for AF & OR are shown in Fig. 1 For the OR, relative ML-TP was equivalent between the manual and automatically generated tracts (ANOVA, p=0.97). Conclusions: Our automated method generated high-quality fibre bundles, with agreement similar to or better than the inter-rater agreement. The most surgically relevant characteristic for ATLR, TP-ML distance for the OR, did not differ significantly. Funding: We are grateful to the Wolfson Foundation and Epilepsy Society for supporting the Epilepsy Society MRI scanner. This work was supported by the NIHR BRC UCLH/UCL High Impact Initiative, Wellcome/EPSRC (203145Z/16/Z), MRC (G0802012, MR/M00841X/1), and Health Innovation Challenge Fund (WT106882).