Comparison of volume- and surface-based magnetic resonance imaging morphometry algorithms in the detection of focal cortical dysplasia.
Authors
Affiliations (9)
Affiliations (9)
- Epilepsy Center, Medical Center, member of the European Reference Network for Rare and Complex Epilepsies EpiCARE, Faculty of Medicine, University of Freiburg, Freiburg, Germany.
- Department of Neuroradiology, Faculty of Medicine, University of Freiburg, Freiburg, Germany.
- Department of Radiology, Medical Physics, Faculty of Medicine, University of Freiburg, Freiburg, Germany.
- Department of Pediatrics and Adolescent Medicine, Medical University of Vienna, member of European Reference Network for Rare and Complex Epilepsies EpiCARE, Vienna, Austria.
- Neurocenter, IT-Service, Faculty of Medicine, University of Freiburg, Freiburg, Germany.
- Department of Neurology and Epileptology, member of European Reference Network for Rare and Complex Epilepsies EpiCARE, Hertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany.
- Department of Neurosurgery, Faculty of Medicine, University of Freiburg, Freiburg, Germany.
- Institute of Neuropathology, Faculty of Medicine, University of Freiburg, Freiburg, Germany.
- Epilepsy Center Kork, Kehl-Kork, Germany.
Abstract
Detection of focal cortical dysplasia (FCD) remains a major challenge in presurgical epilepsy diagnostics. Magnetic resonance imaging (MRI) morphometry increasingly improves lesion detection and postsurgical outcomes. The volume-based Morphometric Analysis Program, version 2018 (MAP18) with integrated artificial neural network and surface-based Multi-centre Epilepsy Lesion Detection (MELD; MELD Graph) MRI morphometry algorithms have demonstrated diagnostic utility, but their performance has not been directly compared within a single cohort of surgically treated patients with histopathologically confirmed FCD. We retrospectively analyzed 3-T MRIs of 32 FCD patients who underwent epilepsy surgery. MAP18 was applied to MP2RAGE (magnetization-prepared 2 rapid gradient echo) sequences and MELD to T1MPRAGE (magnetization-prepared rapid gradient echo) and three-dimensional FLAIR (fluid-attenuated inversion recovery) sequences. Binary lesion masks served as ground truth for evaluating detection accuracy, precision, and recall at cluster and voxel levels. Clinical correlation analysis assessed spatial concordance with seizure onset zone (SOZ) and irritative zone (IZ) from video-electroencephalography (EEG) and postoperative outcome. The cohort included 34 histopathologically proven FCD lesions (91.2% FCD II, 5.8% mild malformation of cortical development, 2.9% unspecified). MELD identified 32 of 34 lesions (94.1%), and MAP18 identified 33 of 34 lesions (97.1%), with concordance in 32 of 34 lesions (94.1%). At the cluster level, MELD showed significantly higher precision (p < .001) and fewer false positive clusters per patient (median 0 vs. 3, p < .001). Voxelwise analysis revealed that MELD demonstrated significantly higher precision than MAP18 (.42 vs. .13, p < .0001) and recall (.44 vs. .23, p < .01). Lobar concordance with SOZ reached 90.6% and with IZ reached 87.5%. Among invasive EEG patients, overlap with SOZ and IZ was 100%. Seizure freedom (Engel I) was achieved in 87.5% of patients. Under standardized high-resolution 3-T MRI, both algorithms achieved high detection rates with strong concordance. Combined application may enhance presurgical lesion detection, with results limited to surgically treated patients.