Abstract
Introduction:
Temporal lobe epilepsy (TLE) is the most common drug-resistant epilepsy in adults. Its hallmark lesion is hippocampal sclerosis which manifests with variably distributed neuronal loss and gliosis across subfields. This lesional entity is apparent on MRI as atrophy, T2 signal changes, and diffusion alterations. Surgery renders the majority of patients seizure-free; yet, individualized outcome predictors remain unknown. We performed a surface-based multivariate integration of volume, T2 intensity, and diffusion parameters across subfields; we evaluated group-level changes with respect to healthy controls and assessed the prognostic yield to predict surgical outcome in single patients.
Methods:
We studied 39 consecutive drug-resistant TLE patients (14 males; 32±9 years; 21/18 LTLE/RTLE). Twenty-one patients underwent surgery; at a follow-up of 42±18 months, 16 (76%) achieved seizure freedom, while 6 (24%) had residual attacks. Hippocampal specimens were of sufficient quality for histopathological analysis in 17 patients, and revealed neuronal loss and gliosis in 10 and isolated gliosis in 7. The control group consisted of 25 age- and sex-matched healthy individuals (12 males; 31±8 years). MRI data were acquired on a 3T Siemens using a 32-channel head coil, including submillimetric T1-weighted (3D-MPRAGE; 0.6×0.6×0.6 mm3 voxels) and T2-weighted MRI (2D-TSE; 0.4×0.4×2.0 mm3 voxels), as well as diffusion-weighted images (twice-refocused 2D-EPI sequence; 2.0×2.0×2.0 mm3 voxels). T1- and T2-weighted images underwent intensity inhomogeneity correction and were linearly registered to the MNI152 template; diffusion images were corrected for head motion and eddy current distortions, and co-registered to the same space. One rater (JKY) manually segmented the subfields cornu. Ammonis (CA) 1-3, CA4 and dentate gyrus (DG), as well as the subiculum. Using our recently described method (Kim et al. 2014), we extracted medial sheets along the central path of a given subfield on which we sampled surface-wise T2 intensity, mean diffusivity, and fractional anisotropy, and calculated local columnar volume changes. Patients were analyzed relative to the epileptogenic lobe (i.e., ipsilateral and contralateral to the seizure focus). Prior to pooling, we normalized measures at a given surface-point using a z-transformation with respect to the corresponding distribution in controls, thereby accounting for
normal inter-hemispheric variations seen in healthy individuals. We used SurfStat for Matlab for statistical analyses (Worsley et al., 2009). Surface-based comparisons assessed anomalies in our patients relative to controls and the relationship to post-surgical outcome. Findings were corrected for multiple comparisons at FWE<0.05. A support vector machine classifier with leave-one-out-cross-validation was employed to predict outcome in single patients. For each patient, we carried out surface-based t-tests between the remaining seizure-free and non-seizure patients to select regions from which features were sampled.
Results:
Compared to controls, patients showed T2 increases across all subfields bilaterally, with most marked effects in ipsilateral CA1-3 and CA4-DG; the latter subfields also showed markedly increased mean diffusivity and decreased anisotropy. Columnar volume was decreased in the ipsilateral CA1-3 and subiculum. Compared to those who achieved seizure freedom, non-seizure free patients displayed bilateral mean diffusivity increases and contralateral columnar volume decreases. A classifier combining volume and mean diffusivity changes accurately predicted surgical outcome in 20/21 (96%) patients.
Conclusions:
Surface-based multivariate integration of morphology, signal and diffusion provides in vivo phenotyping of hippocampal subfield pathology in TLE, and accurately predicts post-surgical outcome.
