Discrimination between neoplastic and nonneoplastic brain lesions by use of proton MR spectroscopy: The limits of accuracy with a logistic regressionmodel

BACKGROUND AND PURPOSE: The most accurate method of clinical MR spectroscop y (MRS) interpretation remains an open question. We sought to construct a l ogistic regression (LR) pattern recognition model for the discrimination of neoplastic from nonneoplastic brain lesions with MR imaging-guided single- voxel proton MRS data. We compared the LR sensitivity, specificity, and rec eiver operator characteristic (ROC) curve area (Az) with the sensitivity an d specificity of blinded and unblinded qualitative MRS interpretations and a choline (Cho)/N-acetylaspartate (NAA) amplitude ratio criterion. METHODS: Consecutive patients with suspected brain neoplasms or recurrent n eoplasia referred for MRS were enrolled once final diagnoses were establish ed by histopathologic examination or serial neurologic examinations, labora tory data, and imaging studies, Control spectra from healthy adult voluntee rs were included. An LR model was constructed with 10 input variables, incl uding seven metabolite resonance amplitudes, unsuppressed brain water conte nt, water line width, and the final diagnosis (neoplasm versus nonneoplasm) , The LR model output was the probability of tumor, for which a cutoff valu e was chosen to obtain comparable sensitivity and specificity. The LR sensi tivity and specificity were compared with those of qualitative blinded inte rpretations from two readers (designated A and B), qualitative unblinded in terpretations (in aggregate) from a group of five staff neuroradiologists a nd a spectroscopist, and a quantitative Cho/NAA amplitude ratio > 1 thresho ld for tumor. Sensitivities and specificities for each method were compared with McNemar's chi square analysis for binary tests and matched data with a significance level of 5%, ROC analyses were performed where possible, and Az values were compared with Metz's method (CORROC2) with a 5% significanc e level. RESULTS: Of the 99 cases enrolled, 86 had neoplasms and 13 had nonneoplasti c diagnoses. The discrimination of neoplastic from control spectra was triv ial with the LR, reflecting high homogeneity among the control spectra, An LR cutoff probability for tumor of 0.8 yielded a specificity of 87%, a comp arable sensitivity of 85%, and an area under the ROC curve of 0,96, Sensiti vities, specificities, and ROC areas (where available) for the other method s were, on average, 82%, 74%, and 0,82, respectively, for readers A and B, 89% (sensitivity) and 92% (specificity) for the group of unblinded readers, and 79% (sensitivity), 77% (specificity), and 0.84 (Az) for the Cho/NAA > 1 criterion. McNemar's analysis yielded significant differences in sensitiv ity (n similar to 86 neoplasms) between the LR and reader A, and between th e LR and the Cho/NAA > 1 criterion. The differences in specificity between the LR and all other methods were not significant (n similar to 13 nonneopl asms), Metz's analysis revealed a significant difference in Az between the LR and the Cho/NAA ratio criterion.