Machine learning-derived multimodal neuroimaging of presurgical target area to predict individual’s seizure outcomes after epilepsy surgery

Objectives: Half of patients who have resective brain surgery for drug-resistant epilepsy have recurrent postoperative seizures. Although several predictors of seizure outcome have been identified, no validated method uses neuroimaging of presurgical target area to predict an individual’s post-surgery seizure outcome. We aimed to develop and validate a machine learning-powered approach to predict an individual’s post-surgery seizure outcome in patients with drug-resistant focal epilepsy.

Methods: One hundred and forty-one patients with drug-resistant focal epilepsy were classified either as having seizure‐free (SZF; Engel class I) or seizure recurrence (SZR; Engel class II through IV) at least 1 year after epilepsy surgery, and data were gathered from January 2016 to August 2018. The presurgical MRI, PET, CT and postsurgical MRI were co-registered for surgical volume of interest (VOI) segmentation. all VOIs were decomposed into nine fixed views, then were inputted the deep residual network (DRN) pretrained on Tiny-ImageNet dataset to extract and transfer deep features. In addition, for improving the representative information, we achieved such non-uniform resolution and sparsity by our mask or VOI based attention operation by enhancing the signal inside the mask and reducing the signal outside of the mask. Multi-Kernel Support Vector Machine (MKSVM) was employed to integrate multiple views of feature sets, and to predict seizure outcomes of the targeted VOIs. Leave-One-Out validation was applied to develop a model for verifying the prediction. In the end, clinical performance using above approach was assessed by calculating accuracy, sensitivity, specificity. Receiver operating characteristic (ROC) curves were generated and the optimal area under the ROC curve (AUC) was calculated as a metric for classifying SZF and SZR.

Results: Application of DRN-MKSVM model based on neuroimaging to predict outcome demonstrated good discrimination. The AUC ranged from 0.799 to 0.952. Importantly, classification performance DRN-MKSVM model using data from multiple neuroimaging showed accuracy of 91.5%, sensitivity of 96.2%, specificity of 85.5% and AUC of 0.95, which were significantly better than any other single modal neuroimaging (all p˂0.05). DRN-MKSVM model with multi-neuroimaging from each individual VOI mask could effectively improve the prediction for classifying SZF from SZR.

Conclusions: DRN-MKSVM using multimodal compared to unimodal neuroimaging from surgical target area accurately predicted postsurgical outcome. It could be conveniently facilitate the preoperative individualized prediction of seizure outcomes in patients who have been judged eligible for epilepsy surgery. This may aid epileptologists in presurgical evaluation by providing a tool to explore various surgical options, offering complementary information to existing clinical techniques. Acknowledgements: This study was supported by the National Natural Science Foundation of China, Grant No. 81801740.

• Download figure
• Open in new tab
• Download powerpoint