Lesion Normalization and Supervised Learning in Post-traumatic Seizure Classification with Diffusion MRI

Area of Science:

• Neuroscience
• Medical Imaging
• Neurology

Background:

• Traumatic brain injury (TBI) can lead to lifelong disabilities, including seizures and post-traumatic epilepsy (PTE).
• Predicting seizure risk in TBI patients is challenging.
• Standard neuroimaging analysis is complicated by physical deformations in moderate-to-severe TBI.

Purpose of the Study:

• To identify TBI patients at risk of developing late seizures.
• To evaluate diffusion-weighted MRI (dMRI) preprocessing strategies for biomarker detection.
• To investigate the utility of fractional anisotropy (FA) features from white matter tracts.

Main Methods:

• Applied four dMRI preprocessing strategies to TBI patient data.
• Included a novel lesion normalization technique for dMRI data.
• Utilized fractional anisotropy (FA) features for seizure prediction.

Main Results:

• The preprocessing pipeline incorporating lesion normalization yielded the best prediction performance.
• Achieved a mean accuracy of 0.819 and a mean area under the curve of 0.785.
• Identified specific white matter tract alterations as potential biomarkers.

Conclusions:

• dMRI with lesion normalization is a promising approach for predicting late seizures after TBI.
• Specific white matter tract alterations can serve as biomarkers for post-traumatic epilepsy risk.
• This method aids in identifying high-risk TBI patients for closer monitoring and intervention.