In a prospective study, patients with a radiologically proven brain tumour underwent diffusion tensor imaging (DTI) prior to definitive diagnosis and treatment. Twenty-eight patients with a histologically proven glioblastoma or metastasis were included in the study. Following the definition of regions of interest, DTI metrics [mean diffusivity (MD) and fractional anisotropy (FA)] were calculated for the tumour volume and the surrounding region of peritumoral oedema. These metrics were then subjected to logistic regression to investigate their ability to discriminate between glioblastomas and cerebral metastases. A cross-validation was performed to investigate the ability of the model to predict tumour. The logistic regression analysis correctly distinguished glioblastoma in 15 of 16 cases (93.8%) and metastasis in 11 of 12 cases (91.7%). Cross-validation resulted in the model correctly predicting 14 of 16 (87.5%) glioblastomas and 10 of 12 (83.3%) metastases studied. MD was significantly higher (p = 0.02) and FA was significantly lower (p = 0.04) within the oedema surrounding metastases than within the oedema around glioblastomas. MD was significantly higher (p = 0.02) within the tumour volume of the glioblastomas. Our results demonstrate that, when DTI metrics from the tumour volume and surrounding peritumoral oedema are studied in combination, glioblastoma can be reliably discriminated from cerebral metastases.
Copyright © 2010 John Wiley & Sons, Ltd.