Quantitative analysis of the morphometric analysis program MAP in patients with truly MRI-negative focal epilepsy

Abstract

Objective: In the presurgical evaluation of epilepsy, identifying the epileptogenic zone is challenging if magnetic resonance imaging (MRI) is negative. Several studies have shown the benefit of using a morphometric analysis program (MAP) on T1-weighted MRI scans to detect subtle lesions. MAP can guide a focused re-evaluation of MRI to ultimately identify structural lesions that were previously overlooked. Data on patients where this additional review after MAP analysis did not reveal any lesions is limited. Here we evaluate the diagnostic yield of MAP in a large group of truly MRI-negative patients.

Methods: We identified 68 patients with MRI-negative focal epilepsy and clear localization of the epileptogenic zone by intracranial EEG or postoperative seizure freedom. High resolution 3D T1 data of patients and 73 healthy controls were acquired on a 3 T scanner. Morphometric analysis was performed with MAP software, creating five z-score maps, reflecting different structural properties of the brain and a patient's deviation from the control population, and a neural network-based focal cortical dysplasia probability map. Ten brain regions were specified to quantify whether MAP findings were located in the correct region. Receiver operating characteristic (ROC) analyses were performed to identify the optimal thresholds for each map.

Results: MAP-guided visual re-evaluation of the original MRI revealed overlooked lesions in three patients. The remaining 65 truly MRI-negative patients were included in the statistical analysis. At the optimal thresholds, maximum sensitivity was 84 %, with 35 % specificity. Balanced accuracy (arithmetic mean of sensitivity and specificity) of the respective maps ranged from 51 % to 60 %, creating three to six times more false positive than true positive findings.

Conclusion: This study confirms that MAP is useful in detecting previously overlooked subtle structural lesions. However, in truly MRI-negative patients, the additional diagnostic yield is very limited.

Keywords: Artificial intelligence; Epilepsy surgery; FCD; Focal cortical dysplasia; MRI; VBM.