Machine Learning Prediction of Surgical Cognitive Outcome Using Multimodal Network MRI
Abstract number :
3.250
Submission category :
5. Neuro Imaging / 5B. Functional Imaging
Year :
2018
Submission ID :
502663
Source :
www.aesnet.org
Presentation date :
12/3/2018 1:55:12 PM
Published date :
Nov 5, 2018, 18:00 PM
Authors :
Karol Osipowicz, Drexel University; Xiaosong He, Thomas Jefferson University; Michael Sperling, Jefferson Comprehensive Epilepsy Center, Thomas Jefferson University; Ashwini Sharan, Thomas Jefferson University; and Joseph Tracy, Thomas Jefferson Universit
Rationale: Developing a quantitative algorithm for the prediction of cognitive outcomes of anterior temporal lobectomy (ATL) for medically intractable mesial temporal lobe epilepsy (mTLE) would represent A significant advance in presurgical decision making. Methods: Utilizing feature selection reduced connectivity matrices derived from presurgical resting-state fMRI (rsfMRI) and diffusion tensor imaging (DTI), we estimated a regression model predicting postsurgical memory scores in 58 mTLE patients who underwent ATL. Results: Our algorithm was able to significantly (p<.001) predict postsurgical memory performance with a least 75% of the variance explained, representing a significant improvement over the clinical predictors alone. Our derived model was able to (with low tolerance) reduce residual error below 20%, and provide a tolerable estimate of actual performance in approximately 85% of our sample. Conclusions: Our results suggest that presurgical MRI can be a robust predictor of postsurgical memory performance, and that in combination with machine learning algorithms, can be used to derive generalizable predictive models. Funding: None