Evaluation of multiple tumour parameters to improve the non-invasive classification of primary brain tumours with 11C-methionine PET

Background: Gliomas demonstrate significant heterogeneity both in terms of outcome and imaging characteristics within histological grades and particularly between histopathological and genetic sub-types. Methionine uptake as measured with positron emission tomography (PET) has been shown to correlate with histological grade but interpretation of the results is difficult unless the histological sub-type is known. We hypothesized that imaging biomarkers of other features known to reflect malignancy including tumour volume, shape and heterogeneity could be measured from methionine PET data to support this interpretation. Method: Methionine PET scans performed in 39 individuals from 2003-4 were identified from the database at the Max-Planck-Institute for Neurological Research in Cologne. Scans were analysed using the ratio to a normal uptake map as previously described. A software application was developed in R to enable standardised, highly reproducible image analysis. Tumour ‘masks’ were defined using a 3D region-growing technique constrained by voxel connectivity. Several statistical measures were recorded for this tumour mask including the volume, characteristics of the border and variability of the methionine uptake including texture measures. Receiver operating characteristic (ROC) analysis was performed for each of these parameters to evaluate accuracy in the discrimination of (1) grade II vs. III/IV and (2) grade III vs. IV tumours. Results: Ten factors were selected including peak methionine uptake, the volume of the tumour mask, measures relating to irregularity of the tumour border / shape, heterogeneity within the tumour volume and the distribution of uptake values within adjacent brain. Each of these features produced significant results on ROC analysis (p=0.000 to 0.010) with the volume of the region of increased methionine uptake being most effective in separating low and high-grade gliomas. Threshold values were defined from the ROC curves to identify results with high specificity for low-grade tumours. 10 of the 13 (77%) grade II tumours analysed exhibited 3 or more of these characteristics as compared to 2 of the 26 (8%) grade III/IV tumours. A score derived from these features demonstrated a greater area under the ROC curve (AUC=0.88) than any of the individual measures in isolation. Conclusion: PET with amino-acid tracers enables multiple tumour measures to be derived from a single diagnostic investigation. Combined evaluation of multiple parameters in an individual tumour may be used to improve accuracy in the non-invasive characterisation of primary brain tumours.