Abstract
Proton magnetic resonance spectroscopy (H-1-MRS) has shown promise in distinguishing recurrent high-grade glioma from posttreatment radiation effect (PTRE). The purpose of this study was to establish objective H-1-MRS criteria based on metabolite peak height ratios to distinguish recurrent tumor (RT) from PTRE. A retrospective analysis of magnetic resonance imaging and H-1-MRS data was performed. Spectral metabolites analyzed included N-acetylaspartate, choline (Cho), creatine (Cr), lactate (Lac), and lipids (Lip). Quantitative H-1-MRS criteria to differentiate RT from PTRE were identified using 81 biopsy-matched spectral voxels. A receiver operating characteristic curve analysis was conducted for all metabolite ratio combinations with the pathology diagnosis as the classification variable. Forward discriminant analysis was used to identify ratio variables that maximized the correct classification of RT versus PTRE. Our results were applied to 205 records without biopsy-matched voxels to examine the percent agreement between our criteria and the radiologic diagnoses. Five ratios achieved an acceptable balance [area under the curve (AUC) ae 0.700] between sensitivity and specificity for distinguishing RT from PTRE, and each ratio defined a criterion for diagnosing RT. The ratios are as follows: Cho/Cr > 1.54 (sensitivity 66%, specificity 79%), Cr/Cho 0.63 (sensitivity 65%, specificity 79%), Lac/Cho ae 2.67 (sensitivity 85%, specificity 58%), Lac/Lip ae 1.64 (sensitivity 54%, specificity 95%), and Lip/Lac > 0.58 (sensitivity 56%, specificity 95%). Application of our ratio criteria in prospective studies may offer an alternative to biopsy or visual spectral pattern recognition to distinguish RT from PTRE in patients with gliomas.