Diagnostic Accuracy of Epilepsy-dedicated MRI with Post-processing

Purpose To evaluate the diagnostic accuracy of epilepsy-dedicated 3 Tesla MRI including post-processing by correlating MRI, histopathology, and postsurgical seizure outcomes. Methods 3 Tesla-MRI including a magnetization-prepared two rapid acquisition gradient echo (MP2RAGE) sequence for post-processing using the morphometric analysis program MAP was acquired in 116 consecutive patients with drug-resistant focal epilepsy undergoing resection surgery. The MRI, histopathology reports and postsurgical seizure outcomes were recorded from the patient’s charts. Results The MRI and histopathology were concordant in 101 and discordant in 15 patients, 3 no hippocampal sclerosis/gliosis only lesions were missed on MRI and 1 of 28 focal cortical dysplasia (FCD) type II associated with a glial scar was considered a glial scar only on MRI. In another five patients, MRI was suggestive of FCD, the histopathology was uneventful but patients were seizure-free following surgery. The MRI and histopathology were concordant in 20 of 21 glioneuronal tumors, 6 cavernomas, and 7 glial scars. Histopathology was negative in 10 patients with temporal lobe epilepsy, 4 of them had anteroinferior meningoencephaloceles. Engel class IA outcome was reached in 71% of patients. Conclusion The proposed MRI protocol is highly accurate. No hippocampal sclerosis/gliosis only lesions are typically MRI negative. Small MRI positive FCD can be histopathologically missed, most likely due to sampling errors resulting from insufficient harvesting of tissue.


Introduction
Around 20-30% of patients with drug-resistant focal epilepsies are considered to be MRI-negative or non-lesional [1][2][3][4][5][6]. Non-lesional means stricto sensu that epileptogenic lesions are by no means made visible on MRI but have a light microscopy correlate; however, one should be aware that many consider MRI also as negative or non-lesional if a lesion was overlooked by a radiologist once.
The most prominent MRI-negative examples are focal cortical dysplasia (FCD) type I with compromise of the radial, tangential or both the radial and tangential orientation of the 6-layered neocortex [7,8]. The abnormal arrangement of cortical neurons in FCD type I should not be invisible on MRI as the cellular density is not changed [9]. Furthermore, the microcolumnar organization of FCD type Ia resembles neuronal radial migration streams during corticogenesis [10,11] and may therefore result from delayed or arrested maturation at mid-gestation [12]. This also holds true for the temporal pole abnormalities associated with hippocampal sclerosis which show a reduced number of axons on diffusion microstructure imaging [13]. The U-fiber layer beneath FCD type I often contains an excessive number of heterotopic neurons leading to a blurring of the graywhite matter junction. These displaced neurons form complex synaptic plexus within the U-fiber layer, some axons of which ascend into the cortex to be integrated into synaptic networks [14]. Accordingly, FCD type I is often accompanied by a regional reduction (lobar hypoplasia/atrophy) and a mild signal increase of the white matter (blurring) making them indirectly visible.
Other examples of MRI-negative histopathological specimens are Palmini et al.'s mild malformations of cortical development (mMCD); however, at least mild malformation of cortical development with excessive white matter neurons (mMCD type 2) produces a blurring of the gray-white matter junction on T2-weighted/FLAIR (Fluid attenuated inversion recovery) sequences [15]. The difficult distinction between no lesion, FCD type I, and mMCD is indicated by the low intrarater agreement documented in a blinded classification of 26 specimens by 8 neuropathologists [16].
Astrogliosis without neuronal loss, called no hippocampal sclerosis, gliosis only may also be missed on MRI. It has been described post-mortem [17] and in surgical specimens [18]. Compared to classical hippocampal sclerosis which accounts for at least 95% of surgical cases, MRI shows absence of pronounced volume loss, only slight T2-weighted signal increase and bilaterality of the disease [19][20][21].
In order to evaluate the accuracy of such an epilepsy-dedicated MRI including post-processing we correlated MRI, histopathology and postsurgical seizure outcomes in a consecutive series of patients with drug-resistant focal epilepsies undergoing resective epilepsy surgery.

Patients
A surgical data base including patients with drug-resistant focal epilepsy undergoing surgery was retrospectively analyzed over a 3-year period. Inclusion criteria were as follows: Epilepsy-dedicated MRI on a Siemens 3 T Prisma scanner (Siemens Healthineers, Erlangen, Germany) including a MP2RAGE sequence. Resective surgery, such as anterior temporal lobectomy, amygdala-hippocampectomy and (extended) lesionectomy. Post-processing using the Morphometric Analysis Program (MAP 18).
Other presurgical evaluations such as video-EEG, stereo electroencephalography (SEEG), and 18 F FDG-PET were considered if available. Patients who had had prior surgery were excluded.

MRI Protocol
The MRI protocol was similar to the International League against Epilepsy (ILAE) recommended harmonized neuroimaging of epilepsy structural sequences (HARNESS) MRI protocol [36]. In addition, a sagittal MP2RAGE sequence was acquired. The study was approved by the local ethics committee (123/20) (Table 1).

Post-processing
The unified images of the MP2RAGE sequence were processed with SPM 12 (http://www.fil.ion.ucl.ac.uk/spm/) running in MATLAB R2014b (MathWorks, Natick, MA, USA). In a first step, DICOM images were converted to Neuroimaging Informatics Technology Initiative (NIfTI) format. NIfTI images were segmented into gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) maps and normalized to the Montreal Neurological Institute (MNI) space. Using the MAP18 software, junction, extension, and thickness images were calculated as described before [37,38]. In addition, these morphometric maps were used as input for an artificial neural network (ANN) trained with MRI data of FCD patients and healthy controls as described elsewhere [39]; however, it should be noted that this ANN has been trained with MPRAGE data as FCD cases with MP2RAGE sequences required for such training are currently not available in sufficient quantity. Table 2 Synopsis of histopathology and MRI for the subgroups malformations of cortical development, hippocampal sclerosis, long-term epilepsy-associated tumors (LEAT), vascular malformations, glial scars, and no lesion. The distribution is similar to the neuropathological series of the European Epilepsy Brain Bank consortium including 9523 patients from 12 European countries [44]. Outcome data were available in 86 patients: The percentages of Engel class IA outcome are related to the histopathology findings  The output of the ANN after classification of all voxels in the unified MP2RAGE image comprises a FCD probability map with values closer to 1 indicating voxels more likely to be dysplastic tissue and values closer to 0 representing non-dysplastic brain tissue or compartments outside of the brain. Finally, all results of MRI post-processing (i.e., morphometric maps and FCD probability map) were inversely normalized (i.e., transferred to native space), the FCD probability maps were co-registered to the MP2RAGE unified images, reconverted to Digital Imaging and Communications in Medicine (DICOM) format and exported to the picture archiving and communication system system (PACS) [27].

Analysis
The MRI reports were retrieved from the PACS system (Dedalus Deep Unity Diagnostic, Bonn, Germany).
Histopathology and outcome reports according to Engel et al. and the International League Against Epilepsy (ILAE) [40,41] were retrieved from the clinical information system system (MeDoc, University Medical Center Freiburg, Germany). Invasive presurgical work-up via stereotactically placed electrodes for SEEG recordings were considered when necessary. Discordant MRI and histopathology findings were reviewed by a neuroradiologist (HU) and a neuropathologist (SD).

Malformations of Cortical Development
The MRI detected abnormalities suggestive of a FCD type II (increased cortical thickness, blurring of the graywhite matter junction, transmantle sign, abnormal gyral/ sulcal pattern) or of a mMCD type II (blurring of the gray-white matter junction) in 38 patients (33%) ( [15,42]; Figs. 2, 3, 4, 5 and 6). On histopathology, a FCD was found in 28 patients and an mMCD type II in 6 patients. In 1 of the 28 patients the FCD was associated with a glial scar, 5 patients without histopathologic abnormalities showed clear MRI abnormalities which were also appreciated on the fully automatic MAP analysis (Fig. 4) and became seizure-free following surgery. Thus, the most likely explanation for missing the histopathological diagnosis in these five patients was a sampling error. Of interest, one FCD type II with an intracortical black line sign was confirmed as FCD type IIA in a second microscopic round after additional resected tissue was con-sidered ( [43]; Fig. 5). Overall, 22 patients with FCD type II had earlier MRI scans stored in the PACS and written reports stored in the medical records: 13 FCD type II (59%) were overlooked (Fig. 2) or not visible at all (Fig. 3).

Hippocampal Sclerosis
Of the patients 31 had hippocampal sclerosis on MRI: in 17 patients, hippocampal sclerosis was the only finding, in 14 patients MRI reports noted additional anterior temporal lobe abnormalities which were described as gray-white matter blurring. Histopathology reports described 14 isolated hippocampal sclerosis and 16 patients with additional temporal lobe abnormalities described as a blurry gray-white matter border (n = 13), FCD type IIA (n = 1), FCD type IIIa (n = 1) or mMCD (n = 1) (Fig. 6).
A hippocampal sclerosis confined to the hippocampal head on MRI was considered negative on histopathology.
A total of three patients were MRI-negative but had histopathologically no hippocampal sclerosis/gliosis only [19]. They were successfully operated on after SEEG elec- Fig. 4 A 14-year-old girl presented with hypermotor seizures since the age of 11. MRI showed a hyperintense left-sided cingulate cortex (a,c,e: arrow). The lesion was found by scrolling through the co-registered MP2RAGE images (b: arrow). The junction parameter map displays the gray-white matter blurring (d: arrow). Histopathology revealed gliotic changes, a gray-white matter blurring, some ectopic white matter neurons, no dysplastic neurons, no balloon cells. MRI after 3 months confirmed that the lesion was resected (f). The patient was seizure-free 12 months following surgery trodes showed ictal spikes in the hippocampus (Engel IA after 12 months).

Long-term Epilepsy-Associated Tumors (LEAT)
A total of 22 patients (19%) had long-term epilepsy-associated tumors (LEAT) of which 14 gangliogliomas and 5 dysembryoplastic neuroepithelial tumors (DNT) were identified on MRI due to the cortical/subcortical cystic elements. The DNTs were separated from gangliogliomas via the multicystic appearance of the glioneuronal element and two pleomorphic xanthoastrocytomas (PXA) were identified due the meningocerebral contrast enhancement. The polymorphous low-grade neuroepithelial tumor of the young (PLNTY) was considered to be most likely a ganglioglioma, although it had no cystic elements and a stalklike extension into the white matter.

Vascular Malformations, Glial Scars
Of the patients six had cavernomas with hemosiderin staining of the brain parenchyma on T2* images, and seven patients had cortical scars. One glial scar on MRI was considered to have an associated FCD IIA on histopathology.

No Lesions
Histopathology was negative in 10 patients suffering from temporal lobe epilepsy ( Table 2) and 4 of the 10 patients were operated on for temporal lobe meningoencephaloceles, so light-microscopic abnormalities were not expected. One had a hippocampal sclerosis on MRI which was not proven by histopathology. With respect to the remaining five patients, all of them were operated on after SEEG electrodes showed seizure onset within the hippocampus and/or the anterior temporal lobe. Only one of them was seizurefree (Engel IA) 12 months following surgery. The FCD shows a funnel-shaped hyperintensity tapering towards the right frontal horn (visible on FLAIR (a) and on T2weighted (b) images: arrow). The initial histopathological diagnosis of a multinodular vacuolating tumor was revised due to the characteristic MRI pattern also highlighted by postprocessing (c). Furthermore, the intracortical black sign is described as a pathognomonic feature of a FCD IIB (d: arrows) In summary, MRI and histopathology were concordant in 101 and discordant in 15 patients. If we consider histopathology as gold standard, sensitivity of MRI is 92.8%, specificity 57.8%, and accuracy 87%; however, we would disregard the fact that five suspected FCD with unequivocal MRI findings which were not confirmed by histopathology and therefore counted as false positives were likely FCD as the patients were seizure-free following surgery (Figs. 4 and 5).

Discussion
Within a 3-year period, we encountered three MRI-negative, histopathology-positive patients undergoing epilepsy surgery: all had no hippocampal sclerosis/gliosis only on histopathology and were successfully operated on after SEEG recordings showed ictal discharges in the hippocampus; however, we also identified six patients with both visible and fully automatically detected MRI lesions that did not have a histopathology substrate. As the five patients with MRI lesions suggestive of a FCD or mMCD (e.g. Figs. 4 and 5) were seizure-free following surgery it is very likely that the small lesions were missed due to sampling errors.
One reason might be that a considerable portion of MRIs are false negative because FCDs are poorly detectable and thus are overlooked. This is especially true for small FCD in the depth of a sulcus. The detection of such lesions is facilitated by high quality MRI scans in conjunction with MRI post-processing, e.g., by means of morphometric analysis. To this end, the Morphometric Analysis Program (MAP18) is employed in our center, which was specifically trained to identify abnormalities such as an abnormal cortical thickness or extension or a blurring of the gray-white matter junction which are summarized in a socalled FCD probability map. The MAP tool distinctly benefits from MP2RAGE images which display the FCD with higher z-scores and larger volumes [25,26]. When the FCD probability maps are inversely normalized (i.e., transferred to native space), co-registered to the original MP2RAGE image, reconverted to DICOM format and exported to the PACS system the FCD can be found very quickly ("within Left-sided hippocampal sclerosis ILAE type 1 (c: blue arrow) with associated gray-white matter blurring (a,b: hollow arrows) in a 38-yearold patient who had fever-related seizures at the age of 8 months. Right-sided hippocampal sclerosis ILAE type 1 (f: blue arrow) in a 50-year-old patient with temporal lobe seizures since the age of 13 years. Coronal FLAIR (d) and post-processed MP2RAGE images (e) showed no associated gray-white matter blurring 1 minute") by scrolling through the co-registered data set [27].
The high rate of epilepsy initially considered to be MRInegative underlines the need to post-process structural MRI scans. Fully automatic MRI analysis may also help to gain confidence that a subtle lesion is a FCD and to proceed with invasive evaluations such as SEEG or directly with surgery.
While hippocampal sclerosis is also clearly visible on MRI, uncertainty exists about the relevance of no hippocampal sclerosis/gliosis only and the additional finding of a gray-white matter blurring of the anterior temporal lobe. There are hints from microscopic examinations and diffusion microstructure imaging that the blurring next to hippocampal sclerosis is a maturation disorder with fewer axons in the anterior temporal lobe [13,45]. This constellation is more likely when the anterior temporal lobe is smaller than its counterpart and clinically supported by the fact that most patients with gray-white matter blurring associated with hippocampal sclerosis had the precipitating event leading to hippocampal sclerosis before the age of 2 years [46]; however, a FCD type I, a FCD type IIa, and a mMCD type II were also described in three patients with gray-white matter blurring underlining that MRI is capable of detecting abnormalities but cannot separate between these different histopathological diagnosis [16].
A total of 10 histopathology-negative patients suffered from temporal lobe epilepsies, 1 had hippocampal sclerosis on MRI, 4 had antero-inferior meningoencephaloceles and 5 were completely MRI-negative. In these patients, epileptogenicity is typically proven via SEEG recordings, in this series only one of five patients became seizure-free.

Limitations
A study to determine the real sensitivity, specificity and accuracy of MRI in epilepsy imaging would randomly include focal epilepsy patients with the problem being that there is no ground truth (histopathology after surgery). This selection bias is reflected in this series by the fact that pa-K tients with a MRI lesion more often underwent surgery than those without; however, the only way to prove whether a MRI lesion has a morphological substrate is to relate it to histopathology. The finding that all MRI-negative and histopathology-negative patients had temporal lobe epilepsies and underwent anterior temporal lobectomy could be due to the fact that is easier to locate an epilepsy syndrome to the anterior temporal lobe than to the other brain lobes, and that patients less frequently have an overlap of the epileptogenic zone with eloquent cortex compared to the central, parietal or frontal neocortex.

Conclusion
True MRI-negative epilepsy defined as light-microscopic abnormalities not being visible on MRI are rare and not in the reported range of 10-30%.
Subtle FCD can be easily missed and represent a major subgroup of MRI false negative epilepsy. They may be highlighted with post-processing tools such as the Morphometric Analysis Program (MAP).
Subtle FCD can even be histopathologically missed. In this constellation, the gold standard for a correct MRI-based diagnosis should be seizure freedom of the patient.
MRI-negative is a difficult term as it includes overlooked lesions, typically FCD type II in the depth of a sulcus. There were three true MRI-negative cases in this surgical series who had no hippocampal sclerosis/gliosis only on histopathology acknowledging the bias that the chance for being operated on is higher when MRI shows a typically epileptogenic lesion.

Funding None
Author Contribution Urbach Horst: study concept, data analysis, preparation of manuscript. Scheiwe Christian, Shah Muskesh J, Nakagawa Julia M: data collection and analysis, proofreading. Heers Marcel, San Antonio-Arce Victoria, Altenmueller Dirk-Matthias, Schulze-Bonhage Andreas: data collection and analysis, presurgical patient selection, proofreading. Huppertz Hans-Juergen, Demerath Theo: postprocessing analysis, proofreading. Doostkam Soroush: data analysis, review of discordant cases, proofreading. Ethical standards The study was approved by the local ethics committee (123/20). Consent for publication: due to the retrospective analysis informed consent of patients was waived.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4. 0/.