Voxel‐based 3D MRI analysis for the detection of epileptogenic lesions at single subject level

To the Editors: With great interest we read the comprehensive and thoughtful review by S.S. Keller and N. Roberts on voxel-based morphometry (VBM) in temporal lobe epilepsy (Keller & Roberts, 2008). Although we share the skepticism of the authors concerning the clinical utility of standard VBM techniques for detecting subtle focal pathology in individuals, we are afraid that some critical statements in this review provoke the misleading impression that this skepticism pertains to all related methods. When discussing the pros and cons of VBM, the authors state that “VBM . . . is not sensitive enough to detect focal pathology in individuals with relatively subtle brain changes, . . . Therefore the utility of VBM during presurgical workup for individual patients is negligible. Variations of the technique have been applied . . . to detect malformations . . . causing epilepsy that are clearly visible on standard T1-weighted images. It is unlikely that standard or optimized VBM techniques will be able to reliably detect subtle malformations that belie visual inspection.” Since some of the technical variations cited above have been developed by authors of this letter, we would like to comment on this statement, with special reference to our own methods. Two aspects deserve discussion: What is the relationship between VBM and “variations of this technique?” What do these methods have in common, and how do they differ from each other? The methods described in our two previous articles (Kassubek et al., 2002; Huppertz et al., 2005), as well as in a more recent study (Huppertz et al., 2008) published in the same volume of Epilepsia as the review, have in common with VBM the use of T1-weighted images, the application of normalization and segmentation algorithms of the statistical parametric mapping (SPM) software, and the comparison with a normal database. They differ, however, with respect to subsequent filters and statistical methods. VBM applies a voxel-wise statistical analysis based on the general linear model and makes statistical inferences using the theory of Gaussian random fields. In contrast, our methods, which we have coined “voxel-based three-dimensional magnetic resonance imaging (3D MRI) analysis” to differentiate them from classical VBM, use only simple descriptive statistics (i.e., calculating mean, standard deviation, and z-scores). We have deliberately refrained from using statistical inferences, since these methods seem to be more appropriate for group comparisons and indeed, as correctly pointed out in the review, suffer either from low sensitivity or low specificity for the detection of subtle malformations in individual patients. Nevertheless, due to automated quantitative analysis of the complete image in three dimensions and the use of special filters in conjunction with an inherent normal database, these “variations of VBM” are able to highlight typical features of cortical malformations, such as blurring of the gray-white matter junction or abnormal distribution of gray matter. With pure visual inspection of conventional MRI however, a direct comparison with norm data is not at hand, and these features can easily be missed. Of course, these methods (so far) do not automatically and unambiguously identify cortical malformations on their own (at least not the very subtle ones). Comparable with a new MRI sequence, they are mainly used to visually direct the reader of the resulting new feature maps to suspicious structural alterations that have to be considered as possible epileptogenic lesions against the background of clinical and electroencephalography (EEG) data. Nonetheless, they help to detect subtle lesions and have proven to be especially helpful in the presurgical workup of epilepsy patients (e.g., either by finding clear-cut lesions that have been overlooked so far, or by generating at least new hypotheses about the epileptogenic lesion, thus guiding the implantation of intracranial electrodes). Several examples from the clinical routine in which the lesion has gone unrecognized in conventional MRI can be found in the articles mentioned above, including lesions located in the temporal lobe. When should we consider a malformation “visible” in conventional images?Keller and Roberts (2008) state that malformations of cortical development detected by variations of the VBM technique are “clearly visible on standard T1-weighted images.” However, to recognize a lesion retrospectively in those image slices that have been selected (or even have been reconstructed from the original data, e.g., with appropriate angulation) to depict the lesion as clearly as possible in a scientific article is a very artificial situation. For clinical reasons, we propose to distinguish lesions that have actually been recognized during individual presurgical decision-making from those signal alterations that are “in principle” retrospectively visible in unprocessed MR images. It seems to be more appropriate to judge a lesion as “visible” if it was actually detected when it came down to clinical decisions. In that sense, there are many examples of cortical malformations in our institution and also other epilepsy centers using these methods that obviously either were not “visible” in conventional MRI or actually have been overlooked in these images despite detailed inspection as a part of presurgical evaluation in a specialized epilepsy surgery center and have only been recognized by help of voxel-based 3D MRI analysis. Furthermore, as a matter of principle, there has to be some kind of brain alteration in the T1 images that, after all, are the basis for subsequent MRI postprocessing. It comes as no surprise that, retrospectively, the lesion underlying the findings in the new feature maps can be detected. In conclusion, we believe that without the use of statistical inferences and the claim of automated lesion detection, variations of the VBM technique can very well be of clinical use during presurgical workup and can also help to detect subtle lesions in individual epilepsy patients. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this letter is consistent with those guidelines. We have no conflicts of interest to disclose.