Multivoxel MR spectroscopic imaging--distinguishing intracranial tumours from non-neoplastic disease.

INTRODUCTION Multi-voxel MR spectroscopic imaging (MRSI) provides chemical metabolite information that can supplement conventional MR imaging in the study of intracranial neoplasia. Our purpose was to use a robust semi-automated spectroscopic analysis to distinguish intracranial tumours from non-neoplastic disease. MATERIALS AND METHODS Twenty intracranial tumours and 15 patients with non-neoplastic disease confirmed on histological examination or serial neuroimaging were studied with 2-dimensional MRSI using point-resolved spectroscopic (PRESS) imaging localisation. Using semi-automated post-processing software, spectra were analysed for peak heights of choline (Cho), creatine (Cr), N-acetyl aspartate (NAA), lactate (Lac) and lipid (Lip). Normalised Cho (nCho) ratios, computed by dividing maximum Cho in the lesion by the normal-appearing brain, were compared between intracranial tumours and non-neoplastic disease. RESULTS Meningiomas displayed homogeneously elevated Cho. Malignant tumours, especially large glioblastoma multiforme, displayed inhomogeneity of metabolites within the tumour. All tumours had elevation of nCho >1 (mean 1.91 +/- 0.65), and non-neoplastic diseases had tumour nCho <1 (mean 0.91 +/- 0.46), which was significantly lower (P <0.05). Two patients with non-neoplastic lesions, one with subacute cerebral infarction and the other with cryptococcoma, had elevated Cho compared to normal tissue (false positive rate 13%). CONCLUSION Using semi-automated MRSI method, a simplified normalised Cho algorithm provides a method to distinguish intracranial tumours from non-neoplastic disease.