Individual Patient Differences in a Pre-Surgical Whole Brain White Matter Connectome Can Predict Post-Surgical Outcome
Abstract number :
962
Submission category :
5. Neuro Imaging / 5A. Structural Imaging
Year :
2020
Submission ID :
2423295
Source :
www.aesnet.org
Presentation date :
12/7/2020 1:26:24 PM
Published date :
Nov 21, 2020, 02:24 AM
Authors :
Walter Hinds, Thomas Jefferson University; Xiaosong He - University of Pennsylvania; Shilpi Modi - Thomas Jefferson University; Kapil Chadhary - Thomas Jefferson University; Ashithkumar Beloor-Suresh - Thomas Jefferson University; Joseph Tracy - Thomas Je
Rationale:
Intractable epilepsy is a neurological disorder that can often benefit from surgical removal of the pathological tissue responsible for generating seizure activity. Employing measures of white matter structural integrity in surgical planning has potential for optimizing intervention strategies on a per patient basis. One possible avenue for testing such potential is the use of the whole-brain structural connectome constructed from diffusion imaging (HARDI) streamline counts.
Method:
White matter tractographic measures were rapidly computed using QSIPrep. Whole brain streamline density counts revealed differences between temporal lobe epilepsy patients (n=35) and healthy controls (n=105). This connectome data was subsequently used to predict 13-18 month outcomes (Engel scale, good versus poor outcome) following surgical procedures (thermal ablation or en bloc temporal lobe resection). Neural network modelling was utilized as this is a method well suited for prediction with smaller sample size data containing many correlated inputs. Results were calculated with a leave-one-out validation method. (see Figure 1)
Results:
Pairwise connections from the whole brain’s connectome was able to discriminate good versus poor outcome significantly better than chance (accuracy=75%; RMSE=.87). The addition of more common clinical predictors to the model (epilepsy duration, age, age of epilepsy onset, surgical procedure) improved classification accuracy to 80%. A post-hoc weight analysis revealed that interhemispheric connections were the main contributors to class prediction, with connections involving the ictal temporal lobe region important as well. The most utilized connection weight was the ""fusiform_L - temporal_pole_mid_L"", involved in the prediction outcome for 33 of 35 patients.
Conclusion:
The potential viability of whole brain structural connectome data as a non-invasive diagnostic for planning neurosurgical intervention is supported by these results. The emphasis in the prediction weights to interhemispheric connections and temporal lobe linkages ipsilateral to seizure onset zone indicates that the pre-surgical integrity of certain specific structural connections may be particularly valuable for the prediction of patient outcomes following surgery.
Funding:
:JIT, NIH/NINDS, R01 NS112816-01
Neuro Imaging