Discrimination between Neoplastic and Nonneoplastic Brain Lesions by Use of Proton MR Spectroscopy: The Limits of Accuracy with a Logistic Regression Model
Jennifer Butzena,
Robert Prosta,
Veerappu Chettya,
Kathleen Donahuea,
Ronald Neppla,
William Bowena,
Shi-Jiang Lia,
Victor Haughtona,
Leighton Marka,
Thomas Kima,
Wade Muellera,
Glenn Meyera,
Hendrikus Krouwera and
Scott Rand
,a
a From the Departments of Radiology, Biophysics (J.B., R.P., V.C., K.D., R.N., W.B., S-J.L., V.H., L.M., T.K., W.M., G.M., H.K., S.R.), Neurosurgery (W.M., G.M.), and Neurology (H.K.), Medical College of Wisconsin, Milwaukee, Wisconsin; the Department of Family Practice (V.K.C.), University of Illinois College of Medicine at Rockford, Rockford, Illinois; Department of Radiology (V.H.), University of Wisconsin Hospitals and Clinics, Madison, Wisconsin; and the Department of Radiology (T.K.), University of Washington Medical Center, Seattle, Washington.
View larger version (7K):
[in a new window]
|
FIG 1. A single-voxel (8-cc) proton spectrum obtained from a mixture of cortex and subcortical white matter within the posterior left frontal lobe of a healthy adult volunteer was acquired with the PRESS pulse sequence (TR = 1500 / TE = 41 / 256 averages) using CHESS water suppression at 0.5 T. The NAA resonance amplitude is approximately twice that of Glx, Cr, Cho and m-Ins
| |
View larger version (49K):
[in a new window]
|
FIG 2. A CT scan of a 61-year-old man with a remote history of radiation therapy for a left glomus jugulare that revealed low attenuation in the left parasagittal occipital lobe was interpreted as compatible with a cortical neoplasm vs ischemia (not shown). An axial T2-weighted image (A) obtained from the same patient reveals thickened, hyperintense cortex with mass effect and adjacent white matter vasogenic edema that were interpreted as most compatible with a neoplasm. An MR spectrum (B) shows diminished NAA, elevated Cho, and no Lip or Lac that was interpreted as neoplastic by an unblinded reader, blinded readers A and B, a Cho/ NAA ratio criterion, and the LR model. Biopsy and histopathologic examination revealed a grade II astrocytoma.
| |
View larger version (56K):
[in a new window]
|
FIG 3. Axial (A) and coronal (B) T2-weighted images (3500/81 [TR/TEeff ]) of a 28-year-old woman with new-onset seizures that depict hyperintense, thickened cortex within the right temporal lobe were interpreted as compatible with a low-grade glioma vs an atypical cortical dysplasia. Postcontrast T1-weighted images showed no lesion enhancement (not shown). An MR spectrum (C) revealing elevated Glx, high Lac (doublet at 1.15 and 1.5 ppm at 0.5 T), diminshed NAA, but no elevation of Cho was interpreted as compatible with a nonneoplastic process by an unblinded reader, blinded reader B (disqualified by A), a Cho/NAA ratio criterion, and the LR model. A subsequent cerebral angiogram was normal. Biopsy and histopathologic examination revealed acute necrotizing vasculitis with recent infarction.
| |
View larger version (9K):
[in a new window]
|
FIG 4. The LR probability of neoplasia for the neoplastic (n=86) and nonneoplastic (n=13) subgroups is presented as a scatter plot. The LR model was constructed with raw metabolite amplitudes.FIG 5. The LR distribution of the probability of neoplasia for the neoplastic (n=86) and nonneoplastic (n=13) subgroups is plotted with an x-axis centile scale. The LR output, the probability of neoplasm, is constrained between 0 and 1 by definition. The 1st centile represents a bin for cases in which LR probability falls between 0 and 0.1, the 2nd centile represents a bin for LR probabilities between 0.1 and 0.2, and so on. For example, 71 neoplastic cases with LR probabilites between 0.9 and 1.0 were distributed into the 10th centile bin. The dashed vertical line over the 8th centile represents the cutoff probability of neoplasm (0.8) chosen to render the sensitivity comparable to the specificity
| |
