Using Gray Matter Probability Maps to Identify Epileptogenic Lesions
Abstract number :
3.197
Submission category :
5. Neuro Imaging / 5B. Structural Imaging
Year :
2016
Submission ID :
196603
Source :
www.aesnet.org
Presentation date :
12/5/2016 12:00:00 AM
Published date :
Nov 21, 2016, 18:00 PM
Authors :
Emily Whitehead, NINDS; Jonathan Scott, Mayo Clinic; Souheil Inati, NINDS; and Sara Inati, NINDS NIH
Rationale: Identification of epileptogenic lesions plays an important role in presurgical evaluation of patients with epilepsy. Identification of focal cortical dysplasias is a particular challenge, as subtle lesions can be easily overlooked, and there can be a variety of findings on structural MRI ranging from obvious FLAIR changes to essentially normal. In these cases, quantitative analysis may add additional information. We implemented a template-free multi-contrast tissue segmentation as described by Saad et al. (1) In this abstract, we evaluate the appearance of epileptogenic lesions using conventional MRI contrasts as well as gray matter probability as a synthetic contrast to aid visual identification of these lesions. Methods: MRIs were obtained on 5 healthy volunteers (2 females, ages 21-40) and 10 patients with epilepsy. Study data were acquired on NIH Clinical Center 3T Philips Achieva MRI Scanners. Imaging included 3D volumetric images (MPRAGE, T2, FLAIR, and Gradient Echo). Tissue classification into gray matter (GM), white matter (WM), and cerebral spinal fluid (CSF) was carried out as described in Saad et al. (1) and Scott et al. (2) Training sets utilized tissue masks created in FreeSurfer from healthy volunteers' images. Classifier training and implementation was performed in Python using logistic regression from the Scikits-learn toolbox. Regions of interest were drawn within each lesion and plotted against normal GM and WM for various contrasts to compare relative intensities. These plots were compared to visual image inspection and GM probability maps. Results: 7 patients with presumed cortical dysplasia (3 pathologically confirmed) were studied. Upon visual inspection, 6 cases had blurring of the gray-white junction on T1 images and 1 patient had cortical thickening only. Using quantitative analysis, intermediate T1 GM-WM intensity was seen in 4 cases. FLAIR and T2 intensity changes were variable on visual and quantitative analysis. 6/7 showed intermediate GM probability using our classifier. In the case with cortical thickening only, the lesion appeared identical to normal GM on visual and quantitative analysis. In cases with a calcified lesion, cavernous angioma, and traumatic encephalomalacia, a variety of intensity patterns were seen on visual and quantitative analysis. All of these lesions were identified as having intermediate GM probabilities using our classifier. Conclusions: These lesions display various intensity patterns across the contrasts in our study. 9/10 lesions were easily identified using our GM probability maps as a synthetic contrast, including lesions that were difficult to identify on the original images. In the future, we will continue to improve this contrast using additional MRI acquisition sequences and textures. References: Saad, Z et al. (2015), Framework for Generating Class Priors From Multi-Contrast Images Without Group Volume Templates, OHBM poster, Honolulu, HI. Scott, J et al. (2015), Clinical Application of a Template-free Framework for Analysis of Multi-contrast Anatomical Images, OHBM poster, Honolulu, HI. Funding: This work was supported by the Intramural Research Program of NINDS.
Neuroimaging