This abstract is a recipient of the Grass Foundation Young Investigator Award This abstract has been invited to present during the Better Patient Outcomes through Diversity Platform poster session This abstract will be presented during the Neuro Imaging Platform poster session
Rationale: Laser interstitial thermal therapy (LITT) is increasingly being used as a less invasive therapy than surgical resection to treat mesial temporal lobe epilepsy (MTLE). However, outcomes are variable, with one-year seizure freedom rates of 36% to 78% (Wu et al., 2019). While MTLE is a focal epilepsy often originating from specific regions of the brain, seizures propagate through long-distance connections that involve several brain structures. Structural connectivity derived from diffusion tensor imaging (DTI) tractography has the potential to inform models predicting seizure outcomes after surgery. The objective of this study is to develop a machine learning (ML) model framework that uses tractography to identify network biomarkers that can predict seizure outcome one year after LITT. Methods: LITT was performed on 26 patients (mean onset age 13.5 +/- 2.2 years, 11 female) using the Visualase laser thermal therapy system (Medtronic) at Harborview Medical Center. Outcomes were classified using the Engel outcome scale. Evidence of mesial temporal sclerosis was recorded based on radiographic findings. Deterministic fiber tracking before and after LITT was performed and individual connectomes were generated for each patient. Local connectivity was measured by clustering coefficient and long-distance connections were measured by number of streamlines between regions. Network change was assessed as a ratio of the postoperative to preoperative measures. Logistic regression (LR) with elastic net regularization was used to reduce connectomic features space and identify network measures before and after LITT associated with seizure freedom. A support vector machine (SVM) with four-fold cross-validation was then applied to this reduced feature set. Post-hoc analysis of network changes across outcome groups was performed using permutation testing. Results: There were no significant differences in demographic and select clinical variables between favorable and unfavorable outcomes one year following surgery (Figfure 1). After LITT, changes in the clustering coefficient of amygdala, hippocampus, and inferior temporal gyrus predicted seizure freedom with 69.6% accuracy, 72.7% sensitivity and 66.7% specificity. Changes in connectivity strength of amygdala, hippocampus, and temporal pole to the isthmus cingulate predict outcomes with 80.6% accuracy, 72.7% sensitivity and 86.7% specificity. There is a greater change in connections between amygdala and isthmus cingulate in the unfavorable compared to favorable outcome group and greater changes in connections between hippocampus, temporal pole, and posterior structures in the favorable compared to unfavorable outcome group (p < 0.01, permutation test; Figfure 1). Conclusions: These findings suggest that changes in connectivity measures of the amygdala and hippocampus are predictive of seizure freedom after LITT and can reveal how well target structures were ablated. Further investigation into the relationship between these network variables and seizure outcomes can improve outcomes by more accurately predicting response to LITT therapy and influencing surgical targeting. Funding: Please list any funding that was received in support of this abstract.: 2019 Radiological Society of North America Medical Student Research Grant
.jpg)
.jpg)