Abstracts

Rationale: Language decline occurs after anterior temporal lobe resection (ATLR) in 30-50% of patients. Damage to white matter bundles has been shown to play a significant, but limited, role in language decline. To investigate this, we perform an in-depth study of white matter connections to individual temporal lobe grey matter regions. We aim to identify the specific changes in white matter connectivity associated with postoperative language decline across language domains.

Methods: We examined data from 54 patients with left temporal lobe epilepsy (TLE) who had left-lateralized language on verbal fluency fMRI and had undergone surgery. Patients’ structural connectomes were derived using deterministic tractography on preoperative diffusion MRI. A modified GIF cortical parcellation [1] was used to parcellate the superior (STG), middle (MTG), inferior (ITG) temporal gyri, and fusiform gyrus (FG) into posterior (p), middle (m), and anterior (a) portions. Post-operative connectomes were estimated by using manually drawn resection masks based on pre-/post-operative 3D-T1 scans which were used as exclusion regions. The change of connections from a region was calculated by dividing the postoperative by the preoperative network. We employed three metrics of language ability: a visual naming task (McKenna Graded Naming test), and semantic, and letter verbal fluency, before and 12 months post-operatively. Pre-operative to post-operative language change was binarized using a reliable change index of -4. An elastic net regularization with varying alpha was used to identify features significantly related to the binary outcomes. Each reduced feature set was entered into a support vector machine incorporating a leave-one-out cross-validation scheme, with the receiving operator characteristic area under the curve (AUC) used to identify the best predictive model. Average accuracy (ACC), sensitivity (SENS), and specificity (SPEC) were also calculated to evaluate the model.

Results: Naming decline was associated with resection of WM connecting to aSTG, aMTG, mMTG, and pFG (AUC = 0.57, ACC = 0.60, SPEC = 0.84, SENS = 0.56). A decline in semantic fluency was associated with resection of WM connecting to aMTG, mMTG, aFG, and pFG (AUC = 0.72, ACC = 0.56, SPEC = 0.70, SENS = 0.73). Letter fluency decline was associated with resection of connections to all MTG, ITG regions, and mFG, pFG regions (AUC = 0.81, ACC = 0.65, SPEC = 0.88, SENS = 0.73).

Conclusions: Preservation of white matter fibers connecting cortical temporal lobe areas in STG, MTG, ITG and FG may be important to avoid post-operative language decline. This method permits a more detailed look into the contributions of white matter connections than full white matter bundle analysis.

[1] Cardosoet al. (2015). Geodesic information flows: spatially-variant graphs and their application to segmentation and fusion. IEEE Trans Med Imaging, 34(9), 1976-1988.

Funding: Please list any funding that was received in support of this abstract.: This work was supported by Epilepsy Research UK (grant number P1904).