1306
Multiparametric Characterization of Focal Cortical Dysplasia Using Three-Dimensional MR Fingerprinting1Epilepsy Center, Neurological Institute, Cleveland Clinic, Cleveland, OH, United States, 2Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 3Biomedical Engineering, Yonsei University, Wonju, Korea, Republic of, 4Quantitative Health Science, Cleveland Clinic, Cleveland, OH, United States, 5Neuropathology, University Hospital Erlangen, Erlangen, Germany, 6Imaging Institute, Cleveland Clinic, Cleveland, OH, United States
Synopsis
Keywords: Epilepsy, MR Fingerprinting, Surgery, Focal cortical dysplasia
Motivation: Focal cortical dysplasia (FCD) is a common pathology in medically intractable focal epilepsy. Detecting and subtyping FCD through visual inspection of conventional MRI can be challenging.
Goal(s): We aimed to develop a multiparametric, quantitative approach for FCD characterization, based on MR fingerprinting (MRF).
Approach: High-resolution 3D MRF scans were performed in 33 epilepsy patients with FCD, 60 normal controls and 26 disease controls. A machine-learning (ML) framework based on MRF was developed to automatically classify FCD from normal cortex and separating FCD subtypes.
Results: MRF-based ML models showed high accuracies, with performances superior to the yields of clinical review.
Impact: Our approach contributes to noninvasive epilepsy presurgical evaluation, as well as an integrated clinical-pathological-imaging understanding of the FCD spectrum.
Introduction
Focal cortical dysplasia (FCD) is a common pathology for medically intractable focal epilepsy. Detecting and subtyping FCD through visual inspection of conventional MRI can be challenging. MR fingerprinting (MRF) is an advanced quantitative MR technique that can acquire multiparametric tissue maps simultaneously1. The quantitative nature of MRF makes it well-suited for reflecting tissue property changes in epileptic lesions. Here, we aim to develop a multiparametric machine learning (ML) approach based on high-resolution three-dimensional MRF for FCD characterization.Methods
Patient recruitment: We included 33 patients with medically intractable focal epilepsy and histologically confirmed FCD who underwent pre-surgical evaluation at the Cleveland Clinic Epilepsy Center. Sixty age-and-gender-matched healthy controls (HCs) and 26 diseased controls (DCs, patients with medically intractable focal epilepsy, and completely nonlesional clinical MRI by official radiology report) were also included.MRI data acquisition: High-resolution 3D whole-brain MRF scans (1 mm3 isotropic voxels) were performed using a Siemens 3T Prisma scanner2 with a 20-channel head coil; field of view=300x300x144 mm3, resolution=1x1x1 mm3, scan time=10 min 24 sec. Dictionary-based reconstruction of the MRF T1 and T2 maps was performed. T1w images were synthesized from the MRF T1 maps.
MRI data processing: The MRF T1 and T2 maps were registered to the Montréal Neurological Institute (MNI) standard space using SyN in Advanced Normalization Tools (ANTs)3. A 3D region of interest (ROI) was manually created for each lesion. ROI-based z-score normalization was performed on MRF T1 and T2 data to minimize bias from lesion location. 2D-level classification between patients and controls was first performed, using mean and standard deviation (SD) of MRF T1 and T2 calculated from the gray matter (GM) and white matter (WM) of the 2D slices as input to a Random Undersampling Boosting ensemble classifier. For FCD subtype classification, we additionally included entropy and uniformity calculated from MRF metrics, as well as the mean and SD values of morphometric maps based on the voxel-based MRI post-processing using the morphometric analysis program(MAP)4,5, namely, the junction, extension, and thickness z-score maps. All features were generated on the 2D slice level. To convert 2D-level classification results to the individual level, the percentage of slices that surpass the adaptive 2D-level threshold (determined by the Youden index from the training dataset) was calculated to indicate the probability of a 3D ROI to be in a given class. Five-fold cross-validation was performed 10 times. Individual-level performance metrics were reported using the mean and SD of the receiver operating characteristic (ROC) curves and the corresponding Area under the curve (AUC) for the 10 trials. Figure 1 details the study workflow.
Histopathological Confirmation: Surgical specimens with immunohistochemical staining were microscopically reviewed by an expert pathologist (IB) for FCD subtypes, using the 2011 ILAE classification guidelines6, with considerations of the critical updates in 20227. For subgroup analysis, we included FCD type IIa, and type IIb for type II. MOGHE8 and mMCD were grouped together as non-type II FCD.
Results
As shown in Figure 2, to classify FCD from HC, the AUC of the ROC curve was 0.974 ± 0.009 (mean ± SD); sensitivity, specificity, and accuracy were 92.4%, 99.3%, and 95.7%, respectively. To classify patients and DCs, the AUC of the ROC curve was 0.973 ± 0.009, with sensitivity of 93.3%, specificity of 95.4%, and accuracy of 94.2%. In comparison, MRI visual review only detected 75.8% (25/33) of the lesions.Classification of FCD type II and non-type-II exhibited AUC of 0.722 ± 0.050, with optimal sensitivity of 76.0%, specificity of 75.4%, and accuracy of 75.8%. Classification of FCD IIa and IIb showed an AUC of 0.874 ± 0.042, with sensitivity of 92.5%, specificity of 83.8%, and accuracy of 89.0%. In comparison, the transmantle sign, the only subtyping sign on conventional MRI, was visible only in 58% (7/12) of the IIb cases.
Discussion
We demonstrated the initial efficacy of a multiparametric machine-learning framework based on high-resolution 3D MRF, to classify FCD from normal cortex, FCD type II from non-type-II, and FCD IIa from IIb. The model performances were superior to the yields of existing approaches (visual inspection assisted by post-processing) in the same cohort, highlighting the efficacy of the quantitative MRI approach. The quantitative image features discovered could provide valuable information for future studies on whole-brain FCD detection based on MRF.Conclusion
Our findings suggest the usefulness of a quantitative imaging approach to augment FCD subtype characterization, which may contribute to the noninvasive pre-surgical evaluation for individuals with epilepsy, as well as an integrated clinical-pathological-imaging understanding of the spectrum of FCD.Acknowledgements
This study was supported by NIH R01 NS109439.References
1. Ma D, Gulani V, Seiberlich N, et al. Magnetic resonance fingerprinting. Nature. 2013;495(7440):187-192.
2. Ma D, Jones SE, Deshmane A, et al. Development of high-resolution 3D MR fingerprinting for detection and characterization of epileptic lesions. J Magn Reson Imaging. 2019;49(5):1333-1346.
3. Avants B, Gee JC. Geodesic estimation for large deformation anatomical shape averaging and interpolation. Neuroimage. 2004;23(SUPPL. 1):S139-150.
4. Huppertz HJ, Wellmer J, Staack AM, et al. Voxel-based 3D MRI analysis helps to detect subtle forms of subcortical band heterotopia. Epilepsia. 2008;49(5):772-785.
5. David B, Kröll-Seger J, Schuch F, et al. External validation of automated focal cortical dysplasia detection using morphometric analysis. Epilepsia. 2021;62(4):1005-1021.
6. Blümcke I, Thom M, Aronica E, et al. The clinicopathologic spectrum of focal cortical dysplasias: A consensus classification proposed by an ad hoc Task Force of the ILAE Diagnostic Methods Commission. Epilepsia. 2011;52(1):158-174.
7. Najm I, Lal D, Alonso Vanegas M, et al. The ILAE consensus classification of focal cortical dysplasia: An update proposed by an ad hoc task force of the ILAE diagnostic methods commission. Epilepsia. 2022;63(8):1899-1919.
8. Schurr J, Coras R, Rossler K, et al. Mild Malformation of Cortical Development with Oligodendroglial Hyperplasia in Frontal Lobe Epilepsy: A New Clinico-Pathological Entity. Brain Pathol. 2017;27(1):26-35.
Figures
