Multi-parametric Characterization of Focal Cortical Dysplasia Using Three-Dimensional MR Fingerprinting
Abstract number :
1.246
Submission category :
5. Neuro Imaging / 5A. Structural Imaging
Year :
2023
Submission ID :
235
Source :
www.aesnet.org
Presentation date :
12/2/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Ting-Yu Su, MS – Cleveland Clinic
Joon Yul Choi, PhD – Epilepsy Center – Cleveland Clinic; Siyuan Hu, BS – Biomedical Engineering – Case Western Reserve University; Xiaofeng Wang, PhD – Quantitative Health Science – Cleveland Clinic; Zheng Ding, MS – Epilepsy Center – Cleveland Clinic; Ken Sakaie, PhD – Imaging Institute – Cleveland Clinic; Ingmar Blümcke, MD – Neuropathology – University Hospitals Erlangen; Hiroatsu Murakami, MD – Epilepsy Center – Cleveland Clinic; Stephen Jones, MD – Imaging Institute – Cleveland Clinic; Imad Najm, MD – Epilepsy Center – Cleveland Clinic; Dan Ma, PhD – Biomedical Engineering – Case Western Reserve University; Zhong Irene Wang, PhD – Epilepsy Center – Cleveland Clinic
Rationale: Focal cortical dysplasia (FCD) is a common pathology in pharmacoresistant focal epilepsy. Detecting and subtyping FCD through visual inspection of conventional MRI can be challenging. We aimed to develop a multivariate machine-learning framework for FCD characterization based on MR Fingerprinting (MRF).
Methods: Thirty-three patients with pathologically-confirmed FCD (13 type I, 8 type IIa, 12 type IIb) were included, as well as 60 age and gender-matched healthy controls (HCs) and 26 diseased controls (DCs, patients with nonlesional clinical MRI by official radiology report). MRI scans were performed using a Siemens 3T Prisma scanner. A 3D whole-brain MRF sequence was used (Ma et al., JMRI 2019). Dictionary-based reconstruction of the MRF T1 and T2 maps was performed. A 3D region of interest (ROI) was manually created for each FCD lesion. FMRIB's Automated Segmentation Tool was applied to segment gray matter (GM) and white matter (WM). MRF maps were registered to Montreal Neurologic Institute space based on the Advanced Normalization Tools. Average templates for MRF T1 and T2 maps were generated from 30 HCs and used for z-score normalization. 3D ROIs were resampled to axial 2D slices. All 2D slices of the same subject were kept together for training and testing. We first performed 2D-level classification between patients and controls, using mean and standard deviation (SD) of MRF values calculated from the GM and WM as input to a Random Undersampling Boosting ensemble classifier. Five-fold cross-validation was performed 10 times to ensure repeatability and reproducibility. For subtype classification, we additionally included entropy and uniformity calculated from MRF values as well as morphometric features calculated from voxel-based MRI post-processing using the morphometric analysis program (David et al., Epilepsia 2021). To convert 2D classification results to individual-level, we used a voting pipeline where the individual-level prediction was determined by the majority of 2D predictions ( >50%). Individual-level performance metrics were reported using mean and SD of the receiver operating characteristic (ROC) curves and the corresponding area under curve (AUC) for the 10 trials. Figure 1 details the study workflow.
Results: As shown in Figure 2, classifying patients and HCs, AUC was 0.994 ± 0.012, with sensitivity of 83.0%, specificity of 99.7% and accuracy of 91.0%. Classifying patients and DCs, AUC was 0.986 ± 0.012, with sensitivity of 81.8%, specificity of 100% and accuracy of 89.8%. In comparison, visual analysis of the MRI only detected 75% (25/33) of the lesions according to official radiology report. Classification of FCD I and II exhibited AUC of 0.752 ± 0.067, with optimal sensitivity of 85.5%, specificity of 57.7% and accuracy of 74.5%. Classification of FCD IIa and IIb showed AUC of 0. 770 ± 0.047, as well as the sensitivity of 80%, specificity of 63.8% and accuracy of 73.5%. In comparison, transmantle sign was visible only in 58% (7/12) of the IIb cases.
Conclusions: Our study revealed efficacy of using 3D MRF to provide automated characterization of FCD, demonstrating the potential to use this noninvasive tool for presurgical evaluation.
Funding: NIH R01 NS109439
Neuro Imaging