Abstracts

1H spectroscopic imaging (SI) of N-acetyl aspartate (NAA) has proven to be highly useful in the identification of neuronal injury in patients with MTLE. Despite its excellent success in seizure lateralization, there are few studies evaluating its accuracy and reproducibility. This issue is of major importance, as NAA is increasingly used for longitudinal studies and therapeutic planning. Previous work reported a range of coefficients of variation (CV, 17-41%) for single pixel measurements of NAA/creatine (NAA/Cr). These variations are due to the large heterogeneitites in metabolite content that normally occur along the hippocampus. To overcome this problem we developed an automated co-registration approach that reduces the study-to-study CV by [sim]50%, reaching 9% for individual pixels and 3.5% over the entire hippocampus. Spectroscopic images were acquired using a single plane spectroscopic imaging sequence, 24x24 encodes 0.64cc/voxel (19min).To provide for reproducible voxel selection and reconstruction, an automated co-registration, selection and reconstruction routine was used. The images were co-registered by maximizing the overlap between two tissue-segmented structural images. Five non-overlapping voxels from each hippocampus (10 per study) were automatically reconstructed by translating along the hippocampal midline with the central voxel placed at the level of the aqueduct. We acquired two hippocampal SI studies in 10 control subjects (mean separation, 34 days), analyzing each dataset two ways. In approach A, we selected 5 voxels from each hippocampus using the SI determined grid following the contours of the hippocampi (no co-registration). In approach B, the automated voxel coregistration and reconstruction was used to evaluate 5 voxels along the mid-line of the hippocampal formation. Approach B demonstrated [sim]50% reduction in CV for individual pixels, 9.0%, versus 17.1% from approach A. Use of the coregistration method B increased the correlation coeffcient between the two studies from R=0.39 to R=0.82 (Figure). Also, approach B reduced the CV of the entire hippocampus (all 5 voxels averaged) by 3 fold, with CVs of 3.5% vs 10.7% (respectively methods B, A).[figure1] Compared to conventional voxel selection routines, these methods reduce the study-to-study CV of Cr/NAA along the hippocampus by 47%. The CV of 9% is [sim]1/5 of mean difference between controls and patients with intractable TLE. This approach should improve longitudinal studies of TLE patients as well as the accuracy of the hippocampal measurement. (Supported by NIH P01-NS-39092, R01-EB000473, Dana Foundation)