Contribution of Quantitative Amygdalar MR FLAIR Signal Analysis for Lateralization of Mesial Temporal Lobe Epilepsy

Abstract

Background and Purpose:

This study evaluates the contribution of an automated amygdalar FLAIR signal analysis for the lateralization of mTLE.

Methods:

Sixty-nine patients (27M, 42F) who had undergone surgery and achieved an Engel class Ia postoperative outcome were identified as a pure cohort of mTLE cases. Forty-six nonepileptic subjects comprised the control group. The amygdala was segmented in T1-weighted images using an atlas-based segmentation. The right/left ratios of amygdalar FLAIR mean and standard deviation were calculated for each subject. A linear classifier (i.e., discriminator line) was designed for lateralization using the FLAIR features and a boundary domain, within which lateralization was assumed to be less definitive, was established using the same features from control subjects. Hippocampal FLAIR and volume analysis was performed for comparison.

Results:

With the boundary domain in place, lateralization accuracy was found to be 70% with hippocampal FLAIR and 67% with hippocampal volume. Taking amygdalar analysis into account, 22% of cases that were found to have uncertain lateralization by hippocampal FLAIR analysis were confidently lateralized by amygdalar FLAIR. No misclassified case was found outside the amygdalar FLAIR boundary domain.

Conclusions:

Amygdalar FLAIR analysis provides an additional metric by which to establish mTLE in those cases where hippocampal FLAIR and volume analysis has failed to provide lateralizing information.

Introduction

Optimal surgical outcomes in cases of mesial temporal lobe epilepsy (mTLE) are dependent upon both imaging and electrographic criteria that seek to reduce uncertainty in lateralization. The distinctive magnetic resonance (MR) imaging features of hippocampal sclerosis (HS) have drawn attention to the hippocampus primarily as a marker site for epileptogenicity. Variance of these features, however, has rendered uncertainty particularly in cases where lateralizing electroencephalographic data has either been entirely absent or discordant. Lack of hippocampal imaging asymmetry is evident in 15–30% of mTLE cases despite clear evidence of ictal origin unilaterally.^1–3 The importance of neighbouring structures, such as the amygdala cannot be overlooked in such circumstances.^4–6

Alterations in amygdalar volume and T2 relaxometry have been noted in cases of mTLE^7–9 and in situ electrographic recordings have confirmed ictal onset here.¹⁰ These features have often been found to accompany changes in the hippocampus as well as in the cerebral cortex, cerebellum and thalamus^11–14 but have also been restricted histopathologically to the amygdala in 10% of patients.^5,6,15–17

Detection of amygdalar MR imaging abnormality may be problematic for its lack of structural distinction.^5,6 Its outline is poorly appreciated because of its small size and insufficient resolution of current standard MR imaging. Prior studies have reported both unilateral and bilateral increased amygdalar T2 relaxation time in cases of mTLE and these often accompany hippocampal sclerosis^13,18,19 making any distinction of the amygdala as a purely epileptogenic source difficult. Moreover, amygdalar T2 relaxation time has often been studied by selecting a region of interest rather than making use of its entire volume.^7,13,18,19 This study introduces fluid-attenuated inversion recovery (FLAIR) mean signal intensity and standard deviation as metrics for studying the amygdala which has been automatically segmented by an atlas-based segmentation technique applied to T1-weighted MR images of each patient. A previous study has shown the lateralization value of hippocampal FLAIR mean and standard deviation, which can characterize changes within the hippocampus caused by varying degrees of pathology. The current work is an extension of that study to the amygdala.

Methods

Subjects

A single institutional archival review, approved by the Henry Ford Health System (Detroit, Michigan, USA) institutional review board, yielded 69 patients (27 M, 42 F; age range 15–66 y, mean±SD = 40±12) operated by a single surgeon (KE) during a 12 year period (1998–2009) (Table 1). Patients had sufficient clinical followup (6.0±3.5, range 4–12 y) to establish surgical outcome specified by Engel Ia classification.²⁰ All patients underwent videoelectroencephalographic inpatient assessment over a minimum three day period, MR imaging, sodium amobarbital study and neuropsychological assessment. Twenty-seven (9 M, 18 F) of the 69 patients (39%) required extraoperative electrocorticography (eECoG) with intracranially implanted electrode arrays. The epileptic focus resided on the right in 26 cases (38%) and on the left in 43 cases (62%). MR images were qualitatively reviewed to render an impression of mesial temporal sclerosis (MTS) defined by reduction of hippocampal volume as determined by a right-left asymmetry on T1-weighted (T1W) coronal images and increased FLAIR MR signal intensity. All surgeries were carried out in the same manner with removal of the inferior temporopolar area along with the uncus, parahippocampal gyrus and hippocampus posteriorly to the tectal plane. There were no associated structural lesions identified in these cases that would distort the mesial temporal anatomy and electrographic features, both extra- and intraoperatively, coincided with a mesial temporal epileptogenicity. Control subjects numbered 46 neurologically intact, nonepileptic individuals without overt lesional pathology (24 M, 22 F; age range 19–61 y, mean±SD = 32±9).

Table 1.

Clinical profiles of patients.

Patient number	Gender	EEG	Side	Age at Surgery	Duration of Epilepsy (y)	Risk Factor	MTS	Pathology
1	F	I/II	L	30	23	CHI(2y);FH	N	gliosis
2	F	I/II	R	34	32	FS(2y);FH	N	-
3	F	I/II	L	50	49	FS(1y);FH	Y	AHS
4	F	I/II	R	31	19	CHI(4,11y)	Y	MDG
5	F	I/II	L	42	38	FH, FS(2y)	Y	gliosis
6	F	I/II	L	54	40	NF1	N	AHS
7	F	I/II	R	40	18	CHI(17,21y)	Y	AHS
8	M	I/II	R	52	50	-	Y	CA
9	F	I/II	R	66	16	meningioma	N	AHS
10	F	I/II	L	37	20	-	Y	gliosis
11	F	I/II	L	43	15	CHI	N	MDG
12	M	I/II	L	32	28	kernicterus	Y	AHS
13	F	I/II	L	54	40	encephalitis	N	gliosis
14	M	I/II	L	38	33	men(8m)	Y	CA, AHS
15	F	I/II	L	47	5	unknown	N	CA
16	F	I/II	L	46	45	FH	Y	CA, AHS
17	F	I/II	L	21	17	none	Y	N/A
18	M	I/II	L	29	20	FS (1-3Y)	Y	paucity of neurons
19	M	I/II	L	46	7	FS	Y	AHS
20	M	I/II	L	39	38	meningitis	N	gliosis; frequent CA
21	F	I/II	R	39	13	unknown	N	—
22	M	I/II	L	54	38	none	Y	AHS
23	M	I/II	L	20	8	none; FS in parents	Y	AHS
24	M	I/II	L	63	31	men(18m)	Y	—
25	F	I/II	R	15	11	FS;FH	Y	AHS
26	F	I/II	R	45	29	unknown	N	CA
27	F	I/II	R	31	8	CHI(16y);FH	N	gliosis
28	M	I	L	31	18	men(6m);CHI(1y)	Y	AHS
29	M	I	L	30	28	FS(2y)	Y	CA
30	F	I	L	57	34	birth trauma	Y	AHS
31	M	I	L	19	18	FCD	N	FCD
32	F	I	L	37	31	FS	Y	non-diagnostic
33	F	I	R	54	15	none	N	non-diagnostic
34	F	I	L	24	18	HypBI	Y	CA, AHS
35	M	I	R	33	22	CHI	Y	AHS/MDG
36	F	I	R	37	30	prematurity	Y	AHS
37	M	I	L	53	16	CHI;FH	Y	AHS
38	M	I	R	36	33	meningitis	Y	gliosis
39	F	I	R	38	26	FS;FH	Y	gliosis
40	F	I	R	34	24	FS;CHI	Y	CA
41	M	I	L	47	48	unknown	Y	AHS
42	M	I	L	40	32	-	Y	AHS
43	F	I	R	32	20	FS	Y	AHS
44	M	I	R	26	21	PNI	Y	AHS
45	F	I	L	56	42	unknown	Y	AHS
46	F	I	L	28	4	aneurysm	Y	gliosis
47	F	I	R	45	34	FS	Y	AHS
48	M	I	L	31	26	none	N	gliosis
49	M	I	L	24	21	prematurity	Y	—
50	F	I	L	48	44	unknown	Y	AHS
51	M	I	L	20	17	CHI	Y	AHS
52	F	I	L	48	12	CHI(34y)	N	AHS
53	F	I	R	34	19	unknown	Y	—
54	F	I	R	50	23	CHI	Y	—
55	F	I	L	39	37	FS	Y	—
56	M	I	L	29	27	unknown	Y	FCD
57	F	I	L	44	39	unknown	Y	AHS
58	F	I	L	60	20	CHI	N (FCD)	FCD
59	M	I	R	57	40	CHI	Y	gliosis
60	M	I	L	56	54	CHI	N	CA
61	F	I	R	45	35	FS	Y	non-diagnostic
62	F	I	R	48	47	FS,CP	Y	gliosis, FCD
63	F	I	L	38	3	lbg(18m),CHI(22y)	N	—
64	F	I	R	29	14	none	Y	non-diagnostic
65	M	I	L	61	59	FS	Y	AHS
66	M	I	R	28	8	CHI	N	—
67	F	I	L	52	31	CHI; fever	Y	AHS
68	M	I	R	49	28	unknown	Y	CA, AHS
69	F	I	L	38	18	men(1.5y)	Y	AHS

Patients are identified by gender, the need for intracranial electrographic study (II) in addition to the preliminary scalp EEG study (I), side of surgery, age at surgery, and duration of epilepsy. The likely initial precipitating event (risk) causing the epilepsy is provided along with the age at which it occurred, if known. The presence or absence of mesial temporal sclerosis (MTS) according to the radiologist’s report establishes the preoperative qualitative interpretation. Histopathology, when available, indicates the main feature of the interpretation. Abbreviations: I, inpatient scalp EEG study; II, extraoperative electrocorticography; AHS, Ammon’s horn sclerosis; CA, corpora amylacea; CHI, closed head injury; CP, complex partial; F, female; FCD, focal cortical dysplasia, FH, family history; FS, febrile seizure; Hipp, hippocampal; HypBI, hypoxic brain injury; L, left; LBG, low blood glucose, M, male; MDG, microdysgenesis; men, meningitis; mgioma, meningioma; NF1, neurofibromatosis type 1; N, no; ND, no diagnosis; PNI, perinatal ischemia; R, right; SG, secondary generalized; SP, simple partial; Vol, volume; Y, yes; y, year.

MR Imaging

Details of the imaging protocols are presented as Supporting Information. In summary, for 40 epilepsy patients and 25 control subjects, coronal T1W and FLAIR images were acquired with a General Electric 1.5T Signa system (GE Medical Systems, Milwaukee WI). For the remaining 29 patients and 21 control subjects, coronal T1W and FLAIR images were acquired with 3T scanners. No patient was imaged by MRI within five days of an ictus that was noticed clinically or identified by the patient.

Image Analysis

A multiatlas-based segmentation technique was used for amygdala segmentation. The atlases consisted of coronal high resolution T1-weighted (T1W) MR images (0.43 mm × 0.43 mm in plane resolution, 1 mm slice thickness, 512×512×172) acquired from 15 neurologically healthy subjects (9 M, 6 F; age 18–50y, mean 32y) along with their manually segmented hippocampus and amygdala using MRIcro (http://www.mccauslandcenter.sc.edu/mricro/). Representative slices of an atlas image along with manually drawn outlines are presented in Fig. 1 A and B. Hippocampal manual segmentation of atlases was performed according to a previously published protocol ²¹ and that for the amygdala using outlines from a published atlas of MR images.²² The T1W atlas images were coregistered with the subject’s T1W MR image using ANTS, a nonrigid coregistration technique.²³ The same transformation was applied to the label volumes. The transformed label volumes were then fused using STAPLE²⁴ to acquire the final labels. All of these operations were performed in the subject’s image space. Fig. 1 C and D displays representative T1W slices of a patient with the segmentation outcome.

Fig. 1.

A and B: Representative T1-weighted (T1W) slices of an atlas image displaying manual amygdalar and hippocampal outlines. C and D: Representative T1W slices of a patient displaying amygdalar and hippocampal outlines generated by multi-atlas based segmentation.

The atlas image resolution provides a clear distinction between the hippocampus and amygdala making it possible to outline the amygdala with higher precision compared to standard T1W images. This made the accuracies of both automated and manual methods comparable as a consequence of the lower image quality and resolution of standard MRI compared to that of the high resolution atlas images. Manual segmentations of the hippocampus^21,25 had been generated by a single investigator (KJ) and verified by another (KE).

For each subject, the T1W MR image was coregistered to the FLAIR MR image using a rigid registration technique (FLIRT)²⁶ based on a correlation ratio. To increase the accuracy of coregistration, nonbrain tissues were eliminated using Brain Extraction Tool (BET)²⁷ and manually corrected whenever needed. The hippocampal and amygdalar labels were mapped to the FLAIR MR images using the registration parameters.

FLAIR mean and standard deviation within each amygdala and hippocampus were calculated in the manner of a previous study.²⁵ To reduce the potential bias caused by the scanners, we corrected the FLAIR intensity inhomogeneity using MRI bias correction in Slicer (http://www.slicer.org/). The right/left ratios of these features were then generated to further reduce the effect of difference between the intensity ranges in 1.5T and 3T scanners. Sample FLAIR images are shown in Fig. 2. Two linear classifiers (i.e., discriminator lines) distinguishing laterality and their boundary domains were generated separately from amygdalar and hippocampal FLAIR MR features. The boundary domain was considered the region in which lateralization of individual cases could not be sufficiently determined. The discriminator line was first determined using Fisher discriminant analysis²⁵ but was also designed to cross the center of the cluster of the 46 control subjects. We rely on the boundary region to provide us a clear separation of definitive right- and left-sided cases. The discriminator line itself is not meant to be the decision-maker of laterality because of the distribution of control cases on either side of it, contained in the boundary region. The boundary domain was determined by two parallel lines with equal distances from the discriminator line representing 95% confidence interval for the control subjects.

Fig. 2.

A and B: Representative fluid-attenuated inversion recovery (FLAIR) slices of patients scanned at 1.5 T and 3 T scanners, respectively. C and D: Representative FLAIR slices of control subjects scanned at 1.5 T and 3 T scanners, respectively.

Although the range and mean value of ages in the control group is slightly different from those of the patient group, further analysis shows insufficient correlation between the extracted features and age (amygdalar FLAIR mean: r= −0.034, p=0.78, amygdalar FLAIR STD: r=0.125, p=0.30, hippocampal FLAIR mean: r=−0.1, p=0.42, hippocampal FLAIR STD: r=−0.047, p=0.70). In addition, as the MR images of patients and control subjects were acquired using scanners with different field strengths, a statistical analysis was run to make certain their features were not biased by field strength. A larger sample size would be required to measure the potentially subtle effect of field strength on classifier parameters.

Hippocampal volume reduction correlates favourably with the side of mTLE.²⁸ A similar impression has been reported for amygdalar volume^14,29 although, in some cases, increased volumes have been reported.^8,9 Amygdalar and hippocampal volumes were analyzed in a manner similar to that in a previous study.^21,25 Volumes were established from segmentation labels and then normalized to (i.e., divided by) the intracranial volume. The accuracy of amygdalar segmentation was judged to be insufficient to detect subtle volumetric asymmetries although it was used for correlation studies as significant volume differences could be detected by automated segmentation. The right-left hippocampal volume difference was used along with thresholding (i.e., with the linear classifier) for lateralization. A boundary domain was generated using volumes of the 46 control subjects. A discriminator line was established as an average of the right-left volume differences in control subjects and this proved to be a close approximation of the y = x line.

Finally, McNemar’s test was used to assess a potential improvement in classification, using the amygdalar FLAIR values to supplement those from hippocampus.

Results

Patients

The mean duration of epilepsy was 26 y (median 26 y; range 3 – 59). The largest etiological grouping, that of closed head injury, was found to manifest an Ammon’s horn sclerosis (AHS) in only 39% of cases, in keeping with prior findings (Table 1).³⁰ Ammon’s horn sclerosis of varying severity and characterized by pyramidal and/or dentate granule cell dropout, gliosis and, in most cases, corpora amylacea was found in 31 cases. Qualitative interpretation of a true AHS may be subject some variance by the degree of change seen absent cell counting. It is possible that some of the interpretations may have been due to sampling given that the entire hippocampus was not submitted for study and there is evidence of a good deal of variability of imaging expression with MTS.²⁵ An imaging diagnosis of MTS coincided with a pathological assessment of AHS in 27/69 (39%) of cases. In a further 13 cases (20%) of MTS, other forms of pathology were indicated (i.e., gliosis, corpora amylacea, cell loss, dysplasia) and, in eight cases, pathology was not available. In 13 of the remaining 18 cases, no MTS was reported but various pathologies were identified although, in only three cases, was an actual AHS reported. These, again, may have amounted to subjective distinctions or a sampling issue. Of the 27 cases requiring eECoG, 11 (41%) harboured a histologically sclerotic hippocampus whereas preoperative imaging declared an MTS in 16 cases (59%) with nine showing an overlap of these attributes.

FLAIR analysis

The ratios (right/left) of amygdalar and hippocampal FLAIR mean and standard deviation are plotted in Figs. 3, 4. No statistically significant difference was found between the amygdalar FLAIR features of the control subjects scanned with the 1.5T and 3T scanners (p=0.19 and p=0.38 for amygdalar FLAIR mean and standard deviation ratios, respectively). Details of the classification results are presented in Table 2. An asymmetric amygdalar FLAIR intensity correctly lateralized in 78% (Fig. 3). In 29% of cases, the lateralization by amygdalar FLAIR asymmetry had higher confidence as defined by placement outside the boundary domain. All subjects with incorrect lateralization in the amygdalar analysis were situated within the boundary domain. Using amygdalar FLAIR asymmetry as a lateralization measure, three patients (7, 18, 23) were correctly lateralized and situated outside the boundary domain but required eECoG. All other patients who had required eECoG were clustered within the boundary domain. The lateralization accuracy of hippocampal FLAIR signal analysis was 87% (Fig. 4) close to a previous estimate of 89% in a cohort of 46 mTLE patients.³¹ With the boundary domain in place, lateralization was possible in 70% for which data was found outside the boundary domain. Eighteen patients (26%) were situated within the boundary domain inside which these features were considered ambiguous.

Fig. 3.

Scatter plot of the mean and standard deviation ratios (right/left) of amygdalar fluid-attenuated inversion recovery (FLAIR) MR signal intensity. Cases are identified by their numerical assignment. Colors red (normal font) and blue (italic font) respectively correspond to left and right surgery sides. The symbol ‘x’ corresponds to the control subjects. SD stands for Standard Deviation.

Fig. 4.

Scatter plots of the mean and standard deviation ratios (right/left) of hippocampal fluid-attenuated inversion recovery (FLAIR) MR signal intensity. Cases are identified by their numerical assignment. Colors red (normal font) and blue (italic font) respectively correspond to left and right surgery sides. The symbol ‘x’ corresponds to the control subjects. SD stands for Standard Deviation.

Table 2.

Classification results.

Patient number	EEG	Side	MTS	Amygdalar FLAIR mean and STD	Hippocampal FLAIR mean and STD	Hippocampal Volume
1	I/II	L	N	Y	Y	N
2	I/II	R	N	Y	Y	Y
3	I/II	L	Y	N	Y	Y
4	I/II	R	Y	Y	Y	N
5	I/II	L	Y	N	Y	Y
6	I/II	L	N	N	Y	Y
7	I/II	R	Y	Y	N	N
8	I/II	R	Y	N	Y	Y
9	I/II	R	N	Y	Y	N
10	I/II	L	Y	Y	Y	Y
11	I/II	L	N	Y	Y	Y
12	I/II	L	Y	Y	Y	Y
13	I/II	L	N	Y	Y	N
14	I/II	L	Y	N	Y	Y
15	I/II	L	N	N	N	N
16	I/II	L	Y	Y	Y	Y
17	I/II	L	Y	Y	Y	Y
18	I/II	L	Y	Y	Y	N
19	I/II	L	Y	Y	Y	Y
20	I/II	L	N	Y	Y	Y
21	I/II	R	N	N	N	N
22	I/II	L	Y	Y	Y	Y
23	I/II	L	Y	Y	Y	N
24	I/II	L	Y	Y	Y	N
25	I/II	R	Y	Y	Y	Y
26	I/II	R	N	N	N	N
27	I/II	R	N	N	N	N
28	I	L	Y	Y	Y	Y
29	I	L	Y	Y	Y	Y
30	I	L	Y	Y	Y	Y
31	I	L	N	Y	Y	N
32	I	L	Y	Y	Y	Y
33	I	R	N	Y	Y	N
34	I	L	Y	Y	Y	Y
35	I	R	Y	Y	Y	N
36	I	R	Y	Y	Y	Y
37	I	L	Y	N	Y	Y
38	I	R	Y	Y	Y	Y
39	I	R	Y	Y	Y	Y
40	I	R	Y	N	Y	Y
41	I	L	Y	Y	Y	Y
42	I	L	Y	Y	Y	Y
43	I	R	Y	Y	Y	Y
44	I	R	Y	Y	Y	Y
45	I	L	Y	Y	Y	Y
46	I	L	Y	N	N	Y
47	I	R	Y	Y	Y	Y
48	I	L	N	Y	N	Y
49	I	L	Y	Y	Y	Y
50	I	L	Y	Y	Y	Y
51	I	L	Y	Y	Y	Y
52	I	L	N	Y	Y	Y
53	I	R	Y	Y	Y	Y
54	I	R	Y	Y	Y	Y
55	I	L	Y	Y	Y	Y
56	I	L	Y	Y	Y	Y
57	I	L	Y	Y	Y	Y
58	I	L	N (FCD)	Y	Y	N
59	I	R	Y	Y	Y	Y
60	I	L	N	Y	Y	Y
61	I	R	Y	N	N	Y
62	I	R	Y	Y	Y	Y
63	I	L	N	Y	Y	Y
64	I	R	Y	N	N	Y
65	I	L	Y	Y	Y	Y
66	I	R	N	N	Y	N
67	I	L	Y	Y	Y	Y
68	I	R	Y	Y	Y	Y
69	I	L	Y	Y	Y	Y

Three columns from Table 1 are repeated here for easier comparison. Patients are identified by the need for intracranial electrographic study (II) in addition to the preliminary scalp EEG study (I) and side of surgery. The presence or absence of mesial temporal sclerosis (MTS) according to the radiologist’s report establishes the preoperative qualitative interpretation. Correct and incorrect classifications are indicated by ‘Y’ and ‘N’, respectively. Cases within the boundary domain are shown by italic Y’s and N’s. Abbreviations: FCD, focal cortical dysplasia; I, inpatient scalp EEG study; II, extraoperative electrocorticography; L, left; MTS, mesial temporal sclerosis; N, no; R, right; Y, yes.

Approximately half the patients who required eECoG were clustered at unity on the hippocampal scatter plot. There was significant correlation between amygdalar and hippocampal mean ratios (r=0.63, p<10⁻⁸). Furthermore, 51 subjects (74%) had similar lateralization based on hippocampal and amygdalar FLAIR mean values. Four patients (cases 7, 23, 48 and 58) out of the 18 demonstrating ambiguity in their hippocampal FLAIR analysis (being situated within the hippocampal FLAIR boundary domain; Fig. 4) were correctly lateralized by definitive amygdalar FLAIR signal asymmetry (i.e., located outside the boundary domain in the correct direction; Fig. 3). The rest of the other 14 patients ambiguously classified by the hippocampal FLAIR boundaries were also ambiguously classified using the amygdalar FLAIR values. Correct classification for 4 out of 4 using the amygdalar definitive region, was statistically significant by McNemar’s test (p=0.046).

In two of the 51 definitively classified patients (48, 58), preoperative imaging did not declare a MTS. Case 7 was reported to have a subtle increase of signal intensity with volume reduction of the overall hippocampal-amygdala complex contralaterally. Interestingly, when the amygdala alone was considered, the preponderant signal intensity change favoured the correct side, suggesting that the contralateral hippocampal signal intensity had been slightly increased even though it was still considered within the boundary domain in our analysis of the hippocampus itself. The majority of patients (16/18; 89%) not declared to have MTS on prior visual assessment were represented very near or within the boundary domain of the amygdalar FLAIR plot (Fig. 3). One patient (64) with MTS was misclassified by hippocampal FLAIR analysis (Fig. 4) and was outside the hippocampal boundary domain although was within the amygdalar boundary domain. This case had visible MTS on the epileptogenic side.

Volume Analysis

Amygdalar volumes in mTLE patients (1219±242 mm³; mean±SD) exceeded those found in control subjects (1153±162 mm³). No statistically significant difference was found between the average amygdalar volumes of the control subjects scanned with the 1.5T and 3T scanners (p=0.11). Patient amygdalar volumes were found comparable for the sides ipsilateral (1216±264 mm³) and contralateral (1220±219 mm³) to that of the epileptogenic side. Hippocampal volumes were 2124±749 mm³ and 2728±503 mm³ for the sides ipsilateral and contralateral, respectively, to that of the epileptogenic side. The lateralization accuracy of hippocampal volumetry (i.e., the right/left hippocampal volume ratio, also specified in Fig. 5 by the y=x+b discriminator line where b corrects the bias) was 75%. With the boundary domain in place, this was reduced to 67%. Twenty-six percent of patients were situated within the boundary domain. Details of the classification results are presented in Table 2. A concordance between amygdalar and hippocampal volume asymmetry was found in 67% of patients. Volumetric distinction was therefore insufficient as a lateralizing metric for the amygdala in mTLE. A number of the misclassified cases in the hippocampal volume scatter plots (4, 7, 27, 31) were found outside the boundary domain (Fig. 5). One of these (27) remained misclassified on the hippocampal FLAIR plot but was inside the boundary domain of the amygdalar FLAIR plot as were two others (4, 31). Two of these (4, 7) had MTS. Case 4 had MTS correlating with the epileptogenic hippocampus while case 7 showed a slight hippocampal volume reduction contralaterally. Amygdalar volumes of the three patients (7, 48, 58) which coincided with the epileptogenic side declared by amygdalar FLAIR analysis were considerably larger on this side (Volumes of epileptogenic side (mm³)/contralateral side (mm³) of 989/884, 1995/1275, 1355/832 for patients 7, 48, 58, respectively).

Fig. 5.

Scatter plots of normalized hippocampal volumes (HVs). Cases are identified by their numerical assignment. Colors red (normal font) and blue (italic font) respectively correspond to left and right surgery sides. The symbol ‘x’ corresponds to the control subjects.

Approximately half the patients who required eECoG were clustered around the y = x+b discriminator line in the hippocampal volume scatter plot (Fig. 5). Again, the majority of patients without MTS were concentrated in the boundary domain of the plot. Case 7 was misclassified by all measures except in the amygdalar FLAIR analysis, in which it was found well outside the boundary domain. No individual amygdalar signal abnormality had been reported for this case by visual analysis. This case was also misclassified by SPECT analysis in a previous study.³¹ A summary of the classification accuracies is presented in Table 3.

Table 3.

Summary of classification accuracies.

Feature set	Accuracy (percentage respectively with and, in brackets, without considering the boundary domain)
Amygdalar FLAIR mean and STD ratios	29 (78)
Hippocampal FLAIR mean and STD ratios	70 (87)
Hippocampal volume	67 (75)

Discussion

FLAIR MR signal intensity is found here to lateralize a predominant amygdalar epileptogenicity in the absence of compelling imaging evidence of hippocampal pathology in a number of mTLE cases. The findings suggest that in those cases of mTLE in which ambiguity exists in lateralizing the epileptogenic focus on the basis of MR imaging attributes assigned to the hippocampus, a detailed MR imaging analysis of the amygdala involving FLAIR may identify a limited area of epileptogenicity sufficient to proceed with surgical resection when confirmed by electroencephalographic (EEG) study.

In this study, amygdalar FLAIR features alone lateralized, with certainty, the side of epileptogenicity in 29% of cases, whereas similar analysis of the hippocampus identified 70% of cases. However, four cases (22%) which could not be lateralized by hippocampal FLAIR analysis (i.e., situated within the boundary domain) were correctly associated with the side of epileptogenicity using amygdalar FLAIR analysis. This finding is in agreement with those of independent amygdalar epileptogenicity established only by in situ recording¹⁰ and surgery.^6,32 The automated segmentations were sufficiently accurate to reveal the role of amygdalar FLAIR analysis in mTLE lateralization. No case found outside the boundary domain in amygdalar FLAIR analysis was misclassified as to the side of epileptogenicity as was the case with hippocampal volumetry and FLAIR analysis. We used imaging data from scanners with different field strengths and corrected the intensity inhomogeneity to make the features less dependent on the scanner and image acquisition protocol.

Using images acquired with different scanners had the disadvantage that the automated method might be affected with an inherent bias, although no statistical bias was observed for this dataset. The two groups of patients (i.e., left-sided and right-sided) had similar portions of patients with data acquired on each scanner. Although scanner variability would potentially decrease lateralization accuracy because of larger overlap in the feature space, our method was able to show the benefit of automated analysis of amygdala in the presence of such variability.

Temporal contusion was not a feature of cases of posttraumatic epilepsy here, according to the pathology reported, particularly in the area of the temporal pole. Moreover, there was no residual hemosiderin deposition within the tissues on MRI that would identify any significant remote injury of this nature. These particular cases appeared to be distinctive manifestations of mild to moderate axonal injury with possible associated ischemic injury. A prior report has reviewed this particular category of mesial temporal epilepsy.³⁰

The absence of bilateral changes may, in part, be explained by the nature of the study in that only pure Engel Ia outcomes were considered. Clinical decisions regarding surgery were also made on the basis of EEG manifestations. Cases in which bilateral manifestations may have occurred could also be contained in the boundary region, similar to controls.

Amygdalar size has been a troublesome aspect of study in mesial temporal epilepsy.^{13,14,18,19,29,33} The mean value of amygdalar volumes for patients with mTLE in this study showed no laterality toward the epileptogenic side. However, it was larger on both sides in patients than was found in controls. Interestingly, in those patients declared to be lateralized solely by amygdalar FLAIR MR signal analysis, the corresponding amygdalar volume was considerably larger on the epileptogenic side compared to that contralaterally. This contrasts with an earlier study that identified the amygdala on the epileptogenic side to be smaller than controls only in those patients with left TLE³³ in the absence of significant asymmetry in TLE patients otherwise. No mention was made of any relation to hippocampal volume or signal intensity on the epileptogenic side in the 24% of patients found to have amygdalar atrophy. Hence, it is unclear whether any of these cases of amygdalar epileptogenicity were exclusive to the amygdala or associated with other mesial temporal structures.

Determination of amygdalar volume may have utility in select cases where a distinct asymmetry exists, particularly when enlarged on the epileptogenic side and when it may be the sole source of epileptogenicity³⁴. Such asymmetric enlargement in association with increased FLAIR MR signal intensity, when comparative hippocampal features of epileptogenicity are absent, suggests that the epileptogenicity may be largely, but not necessarily exclusively, confined to the immediate amygdalar region. Amygdalar connectivity is altered under pathological circumstances³⁵ as in mTLE. Loss of functional connection may serve to isolate the amygdala limiting its epileptogenicity and reducing the recruitment of the neighbouring hippocampus as a nodal point in an epileptogenic network.

Amygdalar change associated with mTLE often coincides with hippocampal change but has been noted to present as an isolated feature in 13% of cases.¹⁴ Standard deviation of amygdalar FLAIR signal intensity was used to assess what was ostensibly felt to be a nonuniform histopathological variation underlying focal epileptogenicity. Characteristic features of cell loss, gliosis and the deposition of corpora amylacea are thought to underlie such variation. Although distinct pathologies such as cortical dysplasia, hamartoma or low grade glioma may account for FLAIR signal alteration, these were not identified microscopically. Signal inhomogeneity may reflect local highly compartmentalized and variable inflammatory expression throughout the amygdala. Certain autoimmune processes^36,37 may be highly selective and in varied stages of development to bring about such variability. The current study indicates that, with quantitative FLAIR MR imaging, a distinction in signal intensity was possible in a greater number of cases than identified previously.

Ratios of right- and left-side features were used in this study to avoid variations in intensity attributable to the specifications in MR imaging units, acquisition conditions and imaging protocols. However, this approach does not account for bilaterally independent epileptogenicity. Perhaps, normalization to the mean value within a reference tissue unaffected by the condition or by imaging artifacts would resolve the issue. Additionally, the amygdala is a relatively small structure.

No single imaging attribute is sufficiently reliable to declare laterality in all cases of mTLE, although the combination of a number of modalities has provided some assurance of success. In the absence of definitive lateralization using the parameters of hippocampal volumetry and FLAIR MR signal attributes, attention must be drawn to neighbouring areas known to be arbiters of epileptogenicity. In those few cases where the amygdala may be such a site, the same quantitative rigour that MR imaging affords must be brought to bear. In 74% of cases, both the amygdala and hippocampus shared imaging attributes that suggested epileptogenicity within the whole and, therefore, would be considered together in the resection volume. The findings here confirm previous impressions that the amygdala as the sole site of epileptogenicity is uncommon. Despite this, a separate amygdalar FLAIR MR analysis may, in a select number of cases, confirm the laterality of a mTLE in situations where such a conclusion was not possible with a similar analysis of the hippocampus.

We have approached the argument of the inclusion of amygdalar FLAIR analysis as a lateralization metric by defining a pure culture of Engel class I outcomes to declare absolute certainty of a mTLE. This would provide the assurance of a definable outcome with this metric alone. An Engel class>1 outcome suggests the possibility of a bilateral temporal epileptogenicity, a temporal plus condition or an extratemporal source. The first two situations might involve a neuroimaging change in the amygdala and the last, presumably not. A study using this approach would require a much greater population of patients to render sufficient power to draw a suitable conclusion.

This report sheds importance upon the added feature of amygdalar FLAIR analysis, moreso than volumetry, as a means of lateralizing a mesial temporal ictal onset where quantitative neuroimaging of the hippocampus itself may fail to provide adequate distinction. A prospective study is required to advance this notion preparatory to creating an algorithmic approach to delineating the extent of the neural substrate constituting the epileptogenic region in the mesial temporal structure.