Abstracts

Rationale: Temporal lobe epilepsy (TLE) is the most common intractable epilepsy in adults, & hippocampal sclerosis (HS) its hallmark lesion. HS is defined by cell loss & gliosis in the subfields. MRI has played a key role in the presurgical evaluation of TLE, however, previous studies have been carried out either on a global hippocampal scale or correspondence between histology & MRI did not employ data registration. In this study, we seek to establish the subfield-specific pathological correlates of hippocampal volume & intensity changes (T1, T2 & DTI) in TLE, & investigate their efficacy in predicting pathology from in-vivo imaging. In-vivo prediction of distinct subfield pathology may lead to more accurate TLE diagnosis & improved patient management.Methods: 15 TLE patients underwent high-resolution pre-op imaging (T1 mapping, T2-w & DTI) on a 3T MRI. After surgery, we automatically quantified neuronal density & gliosis from NeuN & GFAP stains & manually delineated the subfields. To extract MRI parameters we applied our previous pipeline to register the MRI slice corresponding to the cut histology slice. For more robust correlations, we analyzed a select target region encompassing the corresponding MRI slice. Specifically, to model registration & sectioning uncertainty, MRI data adjacent to the corresponding slice were cropped & weighted using a sinc function, giving them a higher weighting. We employed correlation & multiple linear regression analyses to investigate associations between MRI parameters & histological features. We also validated our in-vivo DTI measurements using high-resolution hippocampal ex-vivo DTI (on a 9.4T scanner).Results: A consistent correlation was found between subfield volume & neuronal density, specifically in CA1 (r = 0.91, pfwe < 0.001). MD was correlated with cell density within CA4/DG (r = -0.83, pfwe < 0.001) & T1 correlated in CA4 (r = -0.78, pfwe = 0.006). Higher T2-w related to increased GFAP fraction in CA1 (r = 0.84, pfwe < 0.001). In multiple linear regression analysis, volume, T1 & FA, predicted CA1 % neuronal loss (R2 = 0.90). Volume, T1 & T2 predicted CA2/3 % loss (R2 = 0.96). Loss in CA4 was predicted using volume & MD (R2 = 0.97). T1 of CA2/3 negatively correlated with surgical outcome, i.e., prolonged T1 values relating to better outcomes (r = -0.701, pfwe = 0.012). In-vivo DTI correlated with high-resolution ex-vivo DTI, CA4 (MD r = 0.88, FA r = 0.72) & CA1 (MD: r = 0.58, FA= r = 0.52), validating our in-vivo measurements.Conclusions: This is the first study to investigate the histopathological substrates of volume, T2, quantitative T1 relaxometry while employing high resolution maps & a comprehensive mapping between MRI & pathology. It is also the first direct investigation of histopathological correlates of diffusion metrics in TLE within the hippocampal subfields. We have demonstrated that volume, MD & T1 are sensitive markers for neuronal integrity in the subfields & confirmed that T2 is a marker of gliosis. Moreover, we have shown that in vivo multi-parametric MRI can non-invasively predict subfield pathology, with a precision so far unachievable.