Authors :
Qiangqiang Liu, MD – Ruijin Hospital, Shanghai Jiao Tong University School of Medicine
Haiqing Zhang, PhD – School of Biomedical Engineering, Shanghai Jiao Tong University
Bingyang Cai, PhD – School of Biomedical Engineering, Shanghai Jiao Tong University
Jiwei Li, PhD – School of Biomedical Engineering, Shanghai Jiao Tong University
Hongjia Qi, PhD – School of Biomedical Engineering, Shanghai Jiao Tong University
Jiwen Xu, PhD – Shanghai Ruijin Hospital
Presenting Author: Jie Luo, PhD – School of Biomedical Engineering, Shanghai Jiao Tong University
Rationale:
Ablative surgery targets epileptic foci and networks in drug - resistant temporal lobe epilepsy (TLE) patients. Yet, over 1/3 of patients still experience post - surgery seizures. Accurate prediction of surgical outcome is vital for presurgical planning. Previous research suggested that betweenness centrality could be a good biomarker for integrating temporal structures in outcome prediction. This study explores whole - brain betweenness centrality in its native space (without hemisphere flipping) to find imaging biomarkers based on the presurgical structural network for TLE.
Methods:
In this study, we recruited 46 patients with drug resistant TLE. They underwent structural MRI and diffusion tensor imaging scans both pre- and post- surgery. SEEG RFTC was performed, and patients were followed up for 1 year. Thirty-two age-matched healthy volunteers were also scanned with the same protocol.
Cortical and subcortical gray matter regions were parcellated using the AAL atlas. Ablation masks were manually delineated based on postoperative.
Diffusion data were preprocessed using MRtrix3. Preprocessing included denoising, correction for motion and eddy current distortions, bias field correction, and conversion to MRtrix format. Constrained spherical deconvolution was used to estimate fiber orientation distributions (FODs). Whole-brain probabilistic tractography was then performed using the iFOD2 algorithm, with anatomical constraints and dynamic seeding. Streamline filtering was applied using SIFT2 to improve biological interpretability.
Both fiber number and fractional anisotropy were incorporated in the construction of the structural connectivity matrix. Graph-theoretical metrics were calculated for HC, pre- and post- surgical images, as well as virtual resection scenarios. Group differences features were performed by the student’s t-test and expressed as effect size. We applied LASSO for features selection before using them to in support vector machine (SVM) to predict surgical outcome. Performance was evaluated using leave-one-out cross validation.
Results:
One year after surgery, 50% RTLE, 56% LTLE, and 0% of bilateral TLE had Engel I. RFTC surgery volume had no significant difference between Engel I and II-IV groups. We opt not to flip hemispheres in order to preserve innate anatomical asymmetry in all following analysis. Betweenness Centrality of the right caudate and the right Middle frontal gyrus are significantly elevated in TLE group compared to HC. Top five BC features predictive of outcome include mesial temporal regions: left inferior temporal gyrus and left hippocampus, as well as right putamen, left superior parietal gyrus, left Rolandic operculum, and right mesial superior frontal gyrus, known to participate in seizure propagation networks. The SVM model achieved an AUC of 0.85.
Conclusions:
In summary, this study report Preoperative betweenness centrality features of the unflipped, native-space structural connectome predicted one-year Engel outcome in 46 drug-resistant TLE patients with AUC = 0.85, highlighting the promise of individualized network biomarkers for presurgical planning.
Funding:
Shanghai Municipal Health Commission (No. 202340291)