Lesion Normalization and Supervised Learning in Post-traumatic Seizure Classification with Diffusion MRI.
Md Navid Akbar, Sebastian Ruf, Marianna La Rocca, Rachael Garner, Giuseppe Barisano, Ruskin Cua, Paul Vespa, Deniz Erdoğmuş, Dominique Duncan
Author Information
Md Navid Akbar: Department of Electrical and Computer Engineering, College of Engineering, Northeastern University, Boston, MA 02115, USA.
Sebastian Ruf: Department of Electrical and Computer Engineering, College of Engineering, Northeastern University, Boston, MA 02115, USA.
Marianna La Rocca: USC Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of Southern California, Los Angeles, CA 90033, USA.
Rachael Garner: USC Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of Southern California, Los Angeles, CA 90033, USA.
Giuseppe Barisano: Zilkha Neurogenetic Institute, Keck School of Medicine, University of Southern California, Los Angeles, CA 90033, USA.
Ruskin Cua: Department of Radiology, Keck School of Medicine, University of Southern California, Los Angeles, CA 90033, USA.
Paul Vespa: David Geffen School of Medicine, University of California Los Angeles, Los Angeles, CA 90095, USA.
Deniz Erdoğmuş: Department of Electrical and Computer Engineering, College of Engineering, Northeastern University, Boston, MA 02115, USA.
Dominique Duncan: USC Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of Southern California, Los Angeles, CA 90033, USA.
Traumatic brain injury (TBI) is a serious condition, potentially causing seizures and other lifelong disabilities. Patients who experience at least one seizure one week after TBI (late seizure) are at high risk for lifelong complications of TBI, such as post-traumatic epilepsy (PTE). Identifying which TBI patients are at risk of developing seizures remains a challenge. Although magnetic resonance imaging (MRI) methods that probe structural and functional alterations after TBI are promising for biomarker detection, physical deformations following moderate-severe TBI present problems for standard processing of neuroimaging data, complicating the search for biomarkers. In this work, we consider a prediction task to identify which TBI patients will develop late seizures, using fractional anisotropy (FA) features from white matter tracts in diffusion-weighted MRI (dMRI). To understand how best to account for brain lesions and deformations, four preprocessing strategies are applied to dMRI, including the novel application of a lesion normalization technique to dMRI. The pipeline involving the lesion normalization technique provides the best prediction performance, with a mean accuracy of 0.819 and a mean area under the curve of 0.785. Finally, following statistical analyses of selected features, we recommend the dMRI alterations of a certain white matter tract as a potential biomarker.
