Diagnostic accuracy and interobserver variability of pulsed arterial spin labeling for glioma grading

Abstract

Although pulsed arterial spin labeling (PASL) enables the reliable qualitative grading of brain tumors, its use in quantification for glioma grading may be hampered by the limited interobserver variability associated with low spatial resolution. To assess the interobserver variability and diagnostic accuracy of the relative tumor perfusion signal intensity (rTPS) calculated using PASL in glioma grading. Fifty-eight patients with 61 cerebral astrocytomas underwent conventional MR imaging and PASL. Receiver operating characteristic analyses were used to determine the optimum thresholds for tumor grading. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for identifying high-grade gliomas were also calculated. Cohen's kappa statistic was used to determine the levels of interobserver variability in the quantitative analysis of PASL. The sensitivity, specificity, PPV, and NPV for determining a high-grade glioma with conventional MR imaging were 77.1, 73.1, 79.4, and 70.4%, respectively. A threshold value of 1.28 for rTPS provided a sensitivity, specificity, PPV, and NPV of 82.9, 96.2, 96.7, and 80.6%, respectively. There was a statistically significant difference in the rTPS between low- and high-grade astrocytomas (1.14 vs. 1.47, P<0.05). In the interobserver variability analysis, substantial agreement was obtained for the quantitative rTPS measurement from PASL (kappa =0.72). Quantitative perfusion measurement with PASL can improve the diagnostic accuracy of preoperative glioma grading, as compared to the application of conventional imaging alone. However, the interobserver variability for quantification is substantial.