Abstracts

Rationale: An important issue in epileptology is the detection of structural lesions in MR images. In recent years, automated, voxel-based techniques for structural MRI analysis of T1-weighted 3-dimensional images have been developed. They are mainly applied to clinical studies by group analysis or by comparison of individual images of patients with a healthy control group. Up to now, no VBM-approach has been studied based on Fluid-attenuated inversion recovery (FLAIR) MRI images, since three-dimensional FLAIR-MR-sequences with an isotropic resolution of 1mm are not routinely available, mainly due to the long acquisition time. FLAIR sequence are sensible to water outside the subarachnoidal space, wich is elevated in almost all CNS lesions. This study aimes to develop a tool which can be used on these Flair-based images and be applied to individual patients. This should therefore have a practical application, expanding its use from research to clinical practice. Methods: Three-dimensional FLAIR-images with an isotropic resolution of 1x1x1 mm were acquired in 53 healthy subjects and ten patients with epilepsy (eight of them with temporal lobe epilepsy). Images were corregistered to a 3D-T1 image of the same person and then segmented into grey and white matter. This was done because the spatial resolution of the T1 images surpasses that of the FLAIR, and so a better segmentation is achieved. Voxelwise mean and standard-deviation (SD) were calculated from the original signal intensity values of all healthy subjects in both grey and white matter, and were used to define thresholds upon which structural abnormalities were defined in the patients group (usually signal intensity over 2 SD from the mean). Grey and white matter SD maps can then be shown separately or combined (what is usually not possible by using standard VBM). Results: In the 8 patients with TLE, 4 had a unilateral hippocampus sclerosis (HS), and 4 a limbic encephalitis (LE), where usually the abnormalities are bilateral. On all but two cases (one HS an one LE), all 3D-SD maps showed quite specific abnormalities on the hippocampus, either uni- or bilaterally. In an additional case with a focal cortical dysplasia our method also achieved a localisation of the lesion. The resulting FLAIR-SD maps are displayed overlayed on a on a standard brain for evaluation of results. Conclusions: Voxel-based analysis of MRI sequences is an objective tool for the diagnosis of neurological diseases. Up to now techniques based on T1 fail to identify abnormalities obvious in clinical sequences like FLAIR. Our findings indicate that the three-dimensional FLAIR sequence is well suited for the automated detection of lesions, and that it is applicable to individual cases, as it has been proved in a small selected sample. More extensive and blinded application of this technique will determine its final validity.