Hippocampal subfield segmentation in temporal lobe epilepsy: Relation to outcomes

Objective To investigate the clinical and surgical outcome correlates of preoperative hippocampal subfield volumes in patients with refractory temporal lobe epilepsy (TLE) using a new magnetic resonance imaging (MRI) multisequence segmentation technique. Methods We recruited 106 patients with TLE and hippocampal sclerosis (HS) who underwent conventional T1‐weighted and T2 short TI inversion recovery MRI. An automated hippocampal segmentation algorithm was used to identify twelve subfields in each hippocampus. A total of 76 patients underwent amygdalohippocampectomy and postoperative seizure outcome assessment using the standardized ILAE classification. Semiquantitative hippocampal internal architecture (HIA) ratings were correlated with hippocampal subfield volumes. Results Patients with left TLE had smaller volumes of the contralateral presubiculum and hippocampus‐amygdala transition area compared to those with right TLE. Patients with right TLE had reduced contralateral hippocampal tail volumes and improved outcomes. In all patients, there were no significant relationships between hippocampal subfield volumes and clinical variables such as duration and age at onset of epilepsy. There were no significant differences in any hippocampal subfield volumes between patients who were rendered seizure free and those with persistent postoperative seizure symptoms. Ipsilateral but not contralateral HIA ratings were significantly correlated with gross hippocampal and subfield volumes. Conclusions Our results suggest that ipsilateral hippocampal subfield volumes are not related to the chronicity/severity of TLE. We did not find any hippocampal subfield volume or HIA rating differences in patients with optimal and unfavorable outcomes. In patients with TLE and HS, sophisticated analysis of hippocampal architecture on MRI may have limited value for prediction of postoperative outcome.


| INTRODUC TI ON
Epilepsy is the most common serious neurological disorder.
Refractory temporal lobe epilepsy (TLE) with hippocampal sclerosis (HS) is the most common medically intractable epilepsy condition. 1 Temporal lobe surgery may render between 38 and 60% patients seizure free, depending on the time to postoperative follow-up and definition of seizure freedom. [2][3][4][5] Quantitative magnetic resonance imaging (MRI) techniques provide sensitive surrogate markers of HS. 6 Global hippocampal atrophy is most frequently quantified on T1-weighted (T1w) MRI in patients with TLE. In patients with TLE and HS, hippocampal atrophy has been correlated with various clinical features of the disorder, including age of onset of intractable seizures, duration of epilepsy, and postoperative seizure outcome. [7][8][9] However, other studies have failed to report these associations. 10,11 A potential reason for these discrepancies could be the fact that hippocampal volume alone is not a reliable predictor of post-surgical outcome 11,12 or even of the presence 6 or absence 13 of HS. Indeed, hippocampal volume asymmetry has also been demonstrated in age-matched healthy controls regardless of image presentation during manual measurements. 14 Hippocampal internal architecture (HIA) and variation in hippocampal signal intensity should also be considered alongside volume in context of neuroradiological evaluation. 6,8,13 Signal intensity assessment and semiquantitative HIA ratings are made based on high-resolution coronal MR images, which provide high signal-and contrast-to-noise ratios. HIA ratings can indicate severity and type of HS and may reveal correlations with various clinical features of the disorder. 15,16 Consequently, identifying relationships between clinical features and quantitative characteristics of the hippocampus in TLE is important as they may offer insights into the pathophysiology of the disorder, interindividual patient heterogeneity, and may provide the basis for imaging prognostic markers of treatment outcome.

The International League Against Epilepsy (ILAE) Commission on
Diagnostic Methods have reported three principle patterns of HS based on histopathological analysis. 17 The most common pattern of cell loss, ILAE HS type 1, is manifest as predominant loss of neurons and gliosis in CA1 and CA4 subfields. 17,18 ILAE HS type 2 and 3 are less common patterns of HS, manifest as pathological changes predominantly in CA1 or CA4, respectively. 17,18 Importantly, these patterns of HS appear to be related to various clinical aspects of TLE and may have significance for postoperative prognosis. ILAE HS type 1 is more frequently associated with a history of initial precipitating injuries in early childhood, an early seizure onset and improved seizure outcome after temporal lobe surgery. [17][18][19][20] ILAE HS type 2 and 3 appear to be associated with a later age of onset and a less favorable postoperative outcome, [17][18][19][20] although there are some inconsistencies in these relationships. 21 Given the clinical relevance of regional hippocampal subfield pathology in TLE, it is important to develop and apply MRI approaches that permit assessment of hippocampal subfield structure and volume in this patient group, particularly if such non-invasive imaging measures can be used to predict treatment outcome.
There have been significant advances in the development of MRI techniques for the segmentation and volume estimation of hippocampal subfields. Manual delineation techniques applied to high-field (ie, ≥ 4 Tesla) MRI are the most reliable approaches to identify the approximate location of subfields in individual subjects. [22][23][24] Automated hippocampal subfield approaches applied to high-field MRI have also been described. 25 However, applications of these approaches are constrained by the necessity of nonclinical high-field MRI scanners and the time-inefficient manner of manual tracing. There have therefore been developments of automated hippocampal subfield techniques that can be applied to clinically acquired (ie, ≤ 3 Tesla) MRI data. 26,27 The approach described by Van Leemput et al. (2009) has proved to be particularly popular, given this method's release in context of the freely available Freesurfer software (http://freesurfer.net). 28 We have previously applied this technique to investigate hippocampal subfield alterations in patients with TLE. 29 However, there have been concerns raised with this approach, including reliance on low-resolution T1w images and an imprecise parcellation scheme. 30 Recently, a revised automated hippocampal subfield technique has been introduced that has improved anatomical delineation of the constituent parts of the hippocampus based on multisequence MRI, including standard resolution T1w images and high in-plane resolution T2w images. 31 In a large sample of patients with refractory TLE and HS who underwent conventional T1w and high-resolution T2 short TI inversion recovery (T2STIR) MRI, we have applied this latest approach to investigate whether preoperative in vivo hippocampal subfield analysis had significance for the side of seizure onset, postoperative seizure control, semiquantitative HIA ratings, and other clinical features of TLE.

| Participants
We studied 106 patients with well-characterized mesial TLE and radiological evidence of HS (mean age 40.3 years (SD 13.6); 59 female; 67 with left TLE, 39 with right TLE) who were being evaluated for suitability for neurosurgery at University Hospital Bonn, Germany. Each patient underwent a detailed presurgical program, including comprehensive seizure semiology assessment, MRI, neuropsychological assessment, interictal electroencephalography and if clinically necessary, additional invasive electrophysiological recordings, as reported recently. 10,32 All patients showed evidence of a unilateral temporal lobe seizure onset with concomitant ipsilateral HS. HS was identified by an expert neuroradiologist with considerable experience in lesion diagnosis in epilepsy, which was defined by hippocampal volume loss and internal structure disruption on T1w images, and/or hyperintensities on T2w and FLAIR images. 10 There was no evidence of bilateral HS in any patient; all patients had seizures of presumed unilateral temporal lobe origin, and there was no evidence of a secondary extrahippocampal lesion that may have contributed to seizures. 32 All patients underwent standardized amygdalohippocampectomy and routine diagnostic analysis of resected hippocampal specimens by an experienced neuropathologist. 10 Histological assessment of resected hippocampal specimens revealed that 83% had HS ILAE type I and 17% had HS ILAE type II, no patient had HS ILAE type III. 33 Age of patient, age at diagnosis of epilepsy, duration of epilepsy, history of childhood febrile convulsions, and incidence of secondary generalized tonic-clonic seizures (SGTCS) were recorded for all patients. Patients who underwent temporal lobe surgery (standardized amygdalohippocampectomy) received postoperative follow-up for a period of up to 2 years after surgery and outcome assessment using the ILAE outcome classification system. 34

| MRI analysis
For each patient, we performed quantitative automated segmentation and cortical parcellation of T1w data using Freesurfer version 5.3.0. 28 The standard Freesurfer "recon-all" processing stream was used, which provides surfaces and morphometry data for each subject in addition to gray and white matter segmentations. Automatic labeling and volume estimation of hippocampal subfields were guided by the segmentation of the whole hippocampus (previous step) and performed using the adaptive segmentation technique described by Iglesias et al (2015) in context of the published Freesurfer software version 6 (https://surfer.nmr.mgh.harvard.edu/fswiki/ HippocampalSubfields). Figure 1 shows the anatomical locations of the hippocampal subfields on T1w and T2STIR images in a patient with right TLE after the use of this software module. The protocol coregistered T1w and T2STIR data and used these images simultaneously to generate labels and volumes for the whole hippocampus and 12 hippocampal subfields:

Hippocampal tail
Asymmetry indices for ICV-corrected hippocampal volumes were computed using the previously published formula. 35 Semiquantitative HIA ratings have been shown to be a significant predictor of the laterality of seizure onset in TLE 17,33 and were integrated into image analysis in order to determine if HIA correlates with gross hippocampal and subfield volumes as estimated by Freesurfer version 6. Each T2STIR image slice that depicted the hippocampus was graded with a score of "1" when no internal architecture was perceptible to "4" where excellent internal architecture differentiation could be appreciated. 33 The rater (S.B.E.) was blinded to patient clinical information such as outcome and laterality, and the images were rated on consecutive coronal T2STIR sections in a rostral to caudal direction as described in our previous study. 33 An analysis including resection volumes has been previously performed on this data by the authors and has been fully described. 10

| Statistical analysis
All statistical analyses were performed using MATLAB 2015b.
Group comparison analyses were performed using the unpaired Mann-Whitney U test (data non-normally distributed, P < .05) and included analysis of effects of laterality of epilepsy and postoperative outcome on subfield volume. With respect to postoperative outcome, comparisons were made between patients who attained a postoperative outcome of ILAE 1 (complete seizure freedom) relative to ILAE 2 + (persistent postoperative seizure-related symptoms). 10 Relationships between subfield volume/asymmetry indices and clinical data, including age of onset of epilepsy, epilepsy duration, seizure frequency, and estimated seizure burden, were investigated using Spearman correlation coefficients. Seizure burden was defined as equal to log 10 (frequencyxduration), with the logarithm being applied to accommodate patients with very high-seizure frequency. Correlations were performed corrected for patient age.
Relationships between categorical relationships, including postoperative outcome and sex, side of TLE, and history of childhood febrile/SGTC seizures, were investigated using chi-squared tests of independence. Furthermore, we investigated relationships between HIA ratings and automatically extracted subfield volumes.
Asymmetry indices, gross hippocampal, and subfield volumes were corrected for intracranial volume (ICV), and statistical tests were corrected for multiple comparisons using the false-discovery rate (FDR) procedure.

| RE SULTS
The accuracy of the hippocampal subfield labels was visually checked for all patients. The subfields of one hippocampus in three patients could not be successfully generated. Therefore, analyses were restricted to the 103 with successful reconstructions. Table 1

| Outcome
Of the 103 with successful reconstructions, 76 patients had received amygdalohippocampectomy and postoperative outcome assessment. Of these patients, 41 (54%) patients were seizure free  A breakdown of clinical variables according to outcome is provided in Table 2. An increased number of males was rendered seizure free compared with females (χ 2 = 4.5, P < .05), and right-sided patients were less likely to experience postoperative seizures (χ 2 = 3.7 P = .05). There were no significant differences between outcome groups in incidence of febrile/SGTC seizures, age, age at onset of epilepsy, duration of epilepsy, seizure frequency, or seizure burden.
When all patients were considered together, there were no significant differences in the volume of ipsilateral or contralateral hippocampal subfields between those with postoperative seizure freedom and those with persistent seizure symptoms (P (FDR-corr) > .05; Table 3).
There was also no difference between outcome groups with respect to ICV-corrected hippocampal asymmetry indices (P (FDR-corr) > .05).
No significant differences were observed between outcome groups when patients with left or right TLE were considered separately.
There was also no correlation between the extent of resection and outcome. 10

| Subfield volumes and HIA ratings
Significant correlations were observed between semiquantitative ipsilateral HIA ratings and ipsilateral hippocampal tail (r s = .31;

| D ISCUSS I ON
The present study is the first to use a novel multicontrast approach to improve automated hippocampal subfield segmentation in TLE and to relate these measures to HIA ratings and clinical features.   36,37 in right TLE 35,38 and some studies suggesting equivalence. 39 Just one of these studies 39 has entered the hippocampal asymmetry (right>left) found in healthy controls 14 as a confounding factor during statistical analysis.

| Biological and clinical implications
It is possible that natural cerebral asymmetry of this structure may account for some differences found in patients. 14

| Strengths and limitations
The higher-resolution multicontrast approach used in the present study clearly provided an improved segmentation of hippocampal subfields compared with an automated approach based solely on T1w images. 29

ACK N OWLED G EM ENTS
We would like to acknowledge all participants for taking part in the study and thank the staff involved in data acquisition.