Cortical Measurements from Acute Phase Magnetic Resonance Imaging (MRI) Predict Post Traumatic Epilepsy (PTE) and Functional Outcome after Traumatic Brain Injury (TBI)
Abstract number :
2.305
Submission category :
5. Neuro Imaging / 5A. Structural Imaging
Year :
2025
Submission ID :
426
Source :
www.aesnet.org
Presentation date :
12/7/2025 12:00:00 AM
Published date :
Authors :
Presenting Author: Daniel Jin, BS – Yale School of Medicine
B. Ayvaz, MD – Yale School of Medicine
Justin Wheelock, BA – Yale School of Medicine
Jenna Appleton, PA-C – Yale School of Medicine
Jian Li, PhD – Massachusetts General Hospital
Lawrence Hirsch, MD – Yale University School of Medicine
S. Omay, MD – Yale School of Medicine
Sahar Zafar, MD, MBBS – Massachusetts General Hospital
Aaron Struck, MD – University of Wisconsin, Department of Neurology, Madison, WI
Adrian Dalca, PhD – Massachusetts General Hospital
Brian Edlow, MD – Massachusetts General Hospital
M. Brandon Westover, MD, PhD – Beth Israel Deaconess Medical Center
Emily Gilmore, MD – Yale School of Medicine
Jennifer Kim, MD, PhD – Yale School of Medicine
Rationale: PTE is one of many complications that can hinder recovery after TBI and is the onset of unprovoked seizures after a 7-day acute window. While disruption of the cortex by TBI is known to impact both overall functional recovery and PTE, the quantitative relationship between cortical measurements and these outcomes is less clear. Here we utilize acute phase clinical MRIs to obtain these measurements and compare their use in predicting PTE versus poor 12-month functional outcome after TBI.
Methods: Poor outcome is defined as modified Rankin scale ≥ 3 at 12 months post discharge. We retrospectively identified 131 adult TBI patients with a PTE:non-PTE split of 38:93 and a poor:good outcome split of 44:87. Acute phase T1w images were processed with FreeSurfer v7.4.1 SAMSEG and recon-all to get thickness, thickness standard deviation, and volume measurements for 34 cortical regions in each hemisphere. Lesion masks were made using ScribblePrompt-UNet v0.1.0 to correct for lesions in FreeSurfer processing. FreeSurfer measurements were combined (thicknesses averaged, standard deviations pooled, volumes summed) across hemispheres to create features for prediction. Cox and logistic regression were used to predict PTE and poor outcome respectively. For both prediction targets, significant (p<
Neuro Imaging