Objectives: To assess whether (1)H-MRS may be useful to reinforce the radiological suspicion of PCNSL.
Methods: In this retrospective study, we included 546 patients with untreated brain tumours in which single-voxel spectroscopy at TE 30 ms and 136 ms had been performed. The patients were split into two subgroups: "training set" and "test set." Differences between PCNSL and five other types of intracranial tumours were assessed in the test set of patients using the Mann-Whitney U nonparametric test and cut-off values for pair-wise comparisons defined by constructing receiver operating characteristic curves. These thresholds were used to construct classifiers for binary comparison between PCNSL and non-PCNSL. The performance of the obtained classifiers was assessed in the independent test set of patients.
Results: Significant differences were found between PCNSL and the other groups evaluated. All bilateral comparisons performed in the test set obtained accuracy values above 70 % (71-89 %). Lipids were found to be useful to discriminate between PCNSL and glioblastoma/metastasis at short TE. Myo-inositol resonance was found to be very consistent for discriminating between PCNSL and astrocytomas at short TE.
Conclusions: (1)H-MRS is useful to reinforce diagnostic suspicion of PCNSL on MRI.
Key points: • (1) H-MRS can be used to reinforce the diagnostic suspicion of PCNSL. • Lipids can be used to discriminate between PCNSL and GB/MET. • Myo-inositol resonance can be used to discriminate between PCNSL and astrocytomas.