Characteristics and Surgical Outcomes of Patients With Refractory Magnetic Resonance Imaging–Negative Epilepsies FREE

The rate of successful resective surgery in patients with intractable focal epilepsy has been increasing during the last decades.^1- 2 This is particularly owing to improved brain magnetic resonance imaging (MRI) acquisition and interpretation during presurgical assessment.³ Resectable epileptogenic brain abnormalities have not only become detectable (like Ammon's horn sclerosis [AHS]); in addition, their extension can be more precisely estimated when the resection is planned (as with cortical dysplasias).^4- 6 There are, however, still a proportion of patients seeking presurgical evaluation whose MRIs are not considered to show a lesion potentially causative of chronic epilepsy (referred to hereafter as MRI⁻). These patients have impaired treatment options. Compared with patients with typically epileptogenic structural abnormalities visible on MRI (MRI⁺ patients), MRI⁻ patients are less often offered epilepsy surgery,⁷ and if operations are done, they have an inferior seizure-free outcome.^7- 16 Despite improving MRI technique, MRI⁻ patients account for 18% to 43% of presurgically studied patients with epilepsy.^7,14,16- 17 They can be subdivided into those with distinct epileptogenic lesions on histopathological investigation (histo⁺ patients) and those without such lesions (histo⁻ patients).^{8- 11,13- 17} Little is known as to why epileptogenic lesions are not always detected by preoperative MRI and what the epileptological characteristics of the histo⁻ patients are. To elucidate these 2 points, we analyzed our presurgically studied cohort of MRI⁻ pharmacoresistant patients with epilepsy. The ultimate goal was to explain surgical successes and failures, identify outcome predictors, and draw conclusions about the spatial properties of epileptogenic areas in MRI⁻ patients.

From the prospective database of patients having comprehensive presurgical evaluation for intractable focal epilepsy at Bonn epilepsy center, we identified 1200 patients studied between January 1, 2000, and December 31, 2006.¹⁸ We selected all patients who were diagnosed as MRI⁻. During presurgical evaluation, their MRIs were visually interpreted by experienced neuroradiologists and their findings had been described either as normal or exhibiting abnormalities that are not considered epileptogenic. Patients with the MRI signs of AHS, tumor, malformation of cortical development, vascular malformation, posttraumatic lesions, infarction or bleeding residua, or encephalitis were regarded as MRI⁺. Magnetic resonance imaging was performed on 1.5-T or, since 2005, 3-T systems (Gyroscan ACS-NT, Gyroscan NT-Intera, Gyroscan Intera, Gyroscan 3 T Intera, or Gyroscan 3T Achieva; Philips Medical Systems, Best, the Netherlands) according to an epilepsy protocol that has been described previously.⁶ The MRI⁻ patients who ultimately had epilepsy surgery formed the core group of this study. The following procedures were performed: standard anteromedial temporal lobe resections, tailored temporal lobe resections (taking into consideration individual results from intracranial electroencephalographic [EEG] recordings), selective amygdalohippocampectomies, and extratemporal focal resections, multilobar resections, or functional hemispherectomies. As outcome comparators, we used all MRI⁻ patients who did not have resective surgery and all MRI⁺ patients (subdivided into surgical and nonsurgical patients). All controls came from the same study period. Nonresective surgery, in terms of callosotomy or pure multiple subpial transections, was done in 5 MRI⁻ and 3 MRI⁺ cases. They were excluded from analysis, leaving a total of 1192 patients as the basic study population.

Patients' history, seizure semiology, and EEG data were obtained from hospital files and from original recordings if there were uncertainties. Fluorodeoxyglucose-positron emission tomographies as well as interictal and ictal single-photon emission computed tomographies (SPECT) were recorded and visually interpreted as part of the regular presurgical evaluation at the Department of Nuclear Medicine of the University of Bonn. In some cases, SPECT and MRI data were postprocessed during preoperative assessment using subtraction ictal SPECT coregistered with MRI (SISCOM).¹⁹ Seizure outcome data were obtained from outpatient records of the Department of Epileptology. Patients who were MRI⁻ and did not have follow-up studies were contacted by telephone. The follow-up period was defined as 1 week after surgery or presurgical assessment (in the patients who did not have surgery) until the last available follow-up. Seizure freedom was noted if the patient was free of all seizures including auras during the last year. Only patients with follow-ups more than 6 months later were considered. Whereas “not seizure free” was noted for all patients with ongoing seizures and follow-up of at least half a year, follow-up had to be more than 1 year to be considered seizure free; in patients with a seizure-free follow-up of less than 1 year, “no follow-up” was noted. This was done to avoid overestimation of transient postoperative seizure-free periods.

For the purposes of this study, presurgical MRI data obtained using Avanto 1.5 Tesla or Trio 3.0 Tesla systems (Siemens, Erlangen, Germany; if data from these systems were not available, 3-dimensional T1 data sets from one of the aforementioned Philips systems were used) were postprocessed for detection of subtle signs of focal cortical dysplasia (FCD) using a voxel-based, 3-dimensional, morphometric MRI analysis described elsewhere.^20- 21 In short, this method is based on algorithms of the freely available software for statistical parametric mapping and on additional simple calculations and filters. From a high-resolution T1-weighted 3-dimensional MRI data set, 3 new feature maps were derived that characterize, compared with a normal database, 3 potential features of FCD: the abnormal extension of gray matter into white matter, blurring of the gray matter–white matter junction, and abnormal thickness of the cortical ribbon. Postprocessing was performed by J.W., who at this time was blinded to patients' data. Magnetic resonance images found to be suspicious for presence of FCD were visually reanalyzed by H.U. Postprocessing-negative MRIs were also reread by H.U. for abnormalities potentially overlooked at presurgical assessment. H.U. was not blinded to the patients' clinical and neuropathological data because he participated in the original clinical evaluation and follow-up of most patients.

Presurgical data, MRI postprocessing results and outcome details of the MRI⁻ histo⁻ patients were reviewed in detail to retrospectively provide a hypothesis regarding the spatial extent of their epileptogenic areas.²²

Two-sided Pearson χ² tests and t tests were applied as appropriate. P <.05 was considered significant.

Figure 1 shows the numbers of MRI⁺ and MRI⁻ patients, with the proportions of patients who had surgery and those who were ultimately seizure free. Demographic data, information on presurgical investigations, types of surgery, and follow-up of the groups are given in Table 1. The MRI⁻ patients who had surgery had taken at least 4 antiepileptic drugs without achieving seizure control with tolerable adverse effects; those who did not have surgery were resistant to at least 3 antiepileptic drugs. The number of patients who had surgery was smaller in the MRI⁻ than in the MRI⁺ cohorts (15% vs 73%; P < .001). Reasons for not performing surgery on MRI⁻ patients were lack of clear-cut focus hypothesis after noninvasive monitoring (n = 65; 40%); multiple foci revealed by noninvasive monitoring (n = 10; 6%); multifocal seizure onset documented by invasive monitoring (n = 19; 12%); expected neurological or neuropsychological risks (n = 2; 1%); patient finally decided against surgical approach, in particular, denial of electrode implantation (n = 28;18%); other reason (n = 29; 18%); and more than 1 reason (n = 8; 5%). The proportion of seizure-free MRI⁻ patients who had surgery was lower than in the MRI⁺ group who had surgery (38% vs 66%; P = .003) but higher than in the MRI⁻ patients who did not have surgery (38% vs 16%; P = .008).

The predictive value was estimated for all 29 MRI⁻ patients. The results are given in Table 2. Localizing information was classified as correct positive, correct negative, false positive, false negative, or inconclusive; positive and negative predictive values were calculated (Table 2).

Postprocessing of MRI and seizure semiology both have a relatively low chance of providing localizing information, but if they do, the positive and negative predictive values are high. Other good outcome predictors are surface EEG and SISCOM.

Different types of operative interventions in the MRI⁻ and MRI⁺ groups may account for their different seizure-free outcome rates. Large, less focused standard interventions may offer a higher chance of seizure freedom than small, selective operations. However, in the temporal lobe surgery group, large operations were more common in the MRI⁻ patients than small interventions (eg, selective amygdalohippocampectomy). There was no significant difference between small (focused) and large extratemporal operations (multilobar resections and hemispherectomies) (Table 1). Taken together, there is no indication that the surgical approaches caused the inferior outcome of the MRI⁻ group.

Nine MRI⁻ patients were histo⁺; 3 had FCD type IIB, 2 FCD type IIA,²³ 3 AHS, and 1 cavernoma (Figure 2). The remaining 20 patients had normal or nonspecifically altered histopathological findings. The outcomes of the histo⁺ patients were better than those of the histo⁻ patients (7 of 9 vs 4 of 20 patients seizure free; P = .003). The seizure freedom rate in the MRI⁻ histo⁺ group (78%) is similar to that of MRI⁺ patients (67%; P = .47). Only the 2 patients with FCD IIA did not achieve seizure freedom. There was no significant correlation between the seizure-freedom rate and site of surgery (7 of 22 patients were seizure free after temporal lobe surgery vs 4 of 7 after extratemporal surgery).

Seven of 16 MRI⁻ patients who had 1.5-T scans and 2 of 13 who had 3-T scans were diagnosed with typically epileptogenic lesions on histopathological examination (P = .1). Eight of the 9 histopathological abnormalities could be identified on reassessment of pre–surgically-acquired MRI data. Using voxel-based morphometric MRI postprocessing combined with subsequent visual reinspection of MRIs, all 3 patients with FCD IIB and 1 of the 2 with FCD IIA could be unequivocally detected. Two of the 3 histopathologically diagnosed AHS were still difficult to detect on visual reassessment because of the poor signal to noise ratio of the MRIs; the third AHS was clearly visible and had simply been overlooked during presurgical assessment. Also, the temporolateral cavernoma was readily detectable on reinspection of the presurgical images on which it had been missed. One temporolateral FCD IIA remained undetectable on presurgical MRI data, even with use of MRI postprocessing and visual reassessment, knowing the lesion site and histopathological diagnosis.

Twenty MRI⁻ patients had normal findings or nonspecific alterations on histopathological evaluation. Individual data of these patients and retrospective hypotheses about their epileptogenic areas are given in Table 3 and Table 4. Seventeen had temporal lobe resections. In the remaining 3, extratemporal topectomies were performed (2 in the anterior parts of the medial frontal cortex, 1 in the posterior cortex). One frontal and 3 temporal resections resulted in seizure-free outcomes (Table 3 and Table 4; patients 1-3 and 18). All seizure-free patients were female (P = .02) and had a higher age at onset than the non–seizure-free patients (median age, 29; range 14-31 years vs median, 8; range, 2-19 years; P = .006); epilepsy duration was, however, not different between seizure-free and non–seizure-free patients (median, 20; range, 8-40 years vs median, 18; range, 3-29 years). No seizure-free and 4 non–seizure-free patients had a history that was potentially related to subsequent epilepsy, ie, initial precipitating injuries (3 cases of encephalitides and 1 brain trauma; P = .45).

A case-by-case review of the 20 MRI⁻ histo⁻ patients led to the following hypotheses regarding the spatial extent of their epileptogenic areas according to 5 possible scenarios²⁴:

In this cohort, patients with MRI⁻ epilepsies have a lower chance of having surgery than those with lesions demonstrated by presurgical MRI and, if so, less chance of becoming seizure free. This confirms previous data.^7,25 The seizure-free outcome rate of MRI⁻ patients who had surgery is, on the other hand, better than that of those who did not. This has not been demonstrated before by direct comparison. The MRI⁻ patients for whom surgery was successful demonstrate that the process of multimodal presurgical evaluation may lead to good outcomes in patients with refractory focal epilepsies, even if they do not have an MRI correlate. When choosing a particular test modality, MRI postprocessing, interictal EEG, and semiology have the highest likelihood of providing inconclusive results. If a certain brain area is contemplated as the to-be-resected area, any conclusive test result may be weighted according to the data given in Table 2. Concordant data from MRI postprocessing, semiology, and ictal surface EEG, in that order, are the best predictors of a seizure-free outcome if the planned resection is done. Discordant results of MRI postprocessing, interictal surface EEG, semiology, and SISCOM are the strongest predictors of a non–seizure-free outcome if the contemplated operation is performed. These predictive values and suggestions have, however, several limitations. They are the results of decisions of the center's clinicians, and the values of these tests may vary depending on the location of the pathology (positron emission tomography, for example, is probably better in identifying mediotemporal lobe than extratemporal foci). Both potential biases can hardly be controlled for in this type of retrospective analysis and small sample size. Epilepsies that are MRI^–are not necessarily nonlesional. Thirty percent of MRI⁻ patients in this study and a median proportion of 46% in other series^{8- 11,13- 17} have epileptogenic lesions. In the present cohort, these MRI⁻ histo⁺ patients had a postoperative seizure-free outcome as favorable as that of the MRI⁺ control patients, clearly better than that of the histo⁻ patients who had surgery. Conversely, the proportions of seizure-free patients in the MRI⁺ and MRI⁻ cohorts who did not have surgery are almost identical, suggesting more similarities than dissimilarities regarding the underlying pathology of the epilepsies. It may be expected that previously MRI⁻ histo⁺ patients will be found to be MRI⁺ in the near future, which will increase their chances of proceeding to successful resective epilepsy surgery. Improved MRI acquisition at field strengths of more than 1.5 T and appropriate postprocessing techniques, together with growing experience of MRI evaluators, will likely contribute to this advance.^20,26 This assumption is supported by the results of this series; about three-fourths of the false-negative MRIs (ie, those of MRI⁻ histo⁺ patients) were acquired using 1.5 T systems. Reevaluation by postprocessing and visual reinspection of existing MRI data permitted the detection of underlying brain lesions in all but 1 of 9 such patients. Only 1 temporal lobe FCD IIA remained undetected. Furthermore, the renewed MRI assessment indicated nonresected dysplastic lesions in 2 histo⁻ patients who did not become seizure free after resection of an MRI⁻ histo⁻ area. In both, renewed assessment of presurgical data confirmed that their epileptogenic areas were congruent with the retrospectively identified lesions and had been missed during the original preoperative evaluation. One subsequently had resection after intracranial studies unequivocally confirmed seizure onset in the area, found to be dysplastic during MRI postprocessing.

Presurgical detection of focal lesions by MRI not only guides presurgical evaluation along more effective pathways²⁶; epilepsies caused by such focal lesions appear different in nature from histopathologically nonlesional ones. Seizures related to FCD or AHS arise mostly from strictly circumscribed, and therefore completely resectable, epileptogenic areas (tightly related to the structural abnormalities). In contrast, the major reasons for frequently unsuccessful focal resections in cases without such lesions were multifocality or large extension of the epileptogenic areas (12 of 20, ie, three-fifths). Initial precipitating injuries that cause extended brain damage such as trauma or encephalitis may contribute to widespread epileptogenic areas in those cases. This study further supports the concept that many MRI⁻ patients have poorly localized epileptogenic areas; more than two-thirds of MRI⁻ patients were rejected from surgery after comprehensive evaluation owing to multifocality or poor demarcation of the epileptogenic area. These observations may indicate a more elaborate and extensive presurgical diagnostic procedure (including, for example, more extensive electrode implantations) in this patient group.

Only 3 of 20 histo⁻ patients (15%) had sufficiently restricted epileptogenic areas to become seizure free after resections of typical extension. In this series, seizure-free histo⁻ patients had a relatively high age at epilepsy onset (median, 29 years) and were all female; the meaning and predictive relevance of these features, however, are unclear.

In summary, it is worthwhile to perform presurgical epileptological evaluation of patients with features compatible with monofocal epilepsy, even in the absence of a typically epileptogenic MRI lesion. Patients with concordant noninvasively obtained focus signs should have intracranial EEG and proceed to epilepsy surgery if a monofocal, resectable epileptogenic area is found. At the same time, all efforts should be made to improve the yield of MRI because of the higher effectiveness of presurgical evaluation in MRI⁺ epilepsy patients.

Correspondence: Christian G. Bien, MD, University of Bonn, Department of Epileptology, Sigmund-Freud-Strasse 25, D-53105 Bonn, Germany (christian.bien@ukb.uni-bonn.de).

Accepted for Publication: August 11, 2009.

Author Contributions: The corresponding author had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Bien, Szinay, and Urbach. Acquisition of data: Bien, Szinay, Clusmann, Becker, and Urbach. Analysis and interpretation of data: Bien, Szinay, Wagner, Clusmann, Becker, and Urbach. Drafting of the manuscript: Bien. Critical revision of the manuscript for important intellectual content: Szinay, Wagner, Clusmann, Becker, and Urbach. Administrative, technical, and material support: Clusmann and Becker. Study supervision: Bien, Clusmann, and Urbach.

Financial Disclosure: None reported.

Additional Contributions: The authors would like to thank C. E. Elger, MD, FRCP, University of Bonn, Department of Epileptology and Life and Brain Center, for his kind support and for providing Siemens magnetic resonance imaging data; J. Schramm, MD, for continuous support; and Ms Claudia Ullmann for excellent technical assistance.