Histogram analysis versus region of interest analysis of dynamic susceptibility contrast perfusion MR imaging data in the grading of cerebral gliomas

Abstract

Background and purpose: Histogram analysis can be applied to dynamic susceptibility contrast (DSC) perfusion MR imaging datasets and can be as effective as traditional region-of-interest (ROI) measurements of relative cerebral blood volume (rCBV), an operator-dependent method. We compare the routine ROI method with histogram analysis in the grading of glial neoplasms.

Materials and methods: Ninety-two patients underwent conventional and DSC MR imaging. Routine rCBV (rCBVmax) measurements were obtained from ROIs of the maximal abnormality within the glioma. Histogram analysis rCBVT was performed with an ROI drawn around the maximal tumor diameter. Spearman rank correlations measured associations among glioma grade, rCBVmax, and histogram measures. Mann-Whitney tests compared grade with respect to rCBV and histogram measures. Logistic regression and McNemar test compared the utility of rCBVmax and histogram measures for detecting high grade gliomas.

Results: Routine rCBVmax analysis showed significant correlation with grade (r = 0.734, P < .001). Histogram rCBVT metrics showed significant correlation with grade (P < .008); the 3 highest were rCBVT SD, SD50, and mean25 (r = 0.718, 0.684, and 0.683, respectively). Grade could be predicted by rCBVmax (P < .001) as well as rCBV(T) (P < .008). Three rCBVT histogram measures (SD, SD25, and SD50) detected high-grade glioma with significantly higher specificity than rCBVmax when the diagnostic tests were constrained to have at least 95% sensitivity.

Conclusion: rCBVT histogram analysis is as effective as rCBVmax analysis in the correlation with glioma grade. Inexperienced operators may obtain perfusion metrics using histogram analyses that are comparable with those obtained by experienced operators using ROI analysis.

Fig 1.

Sample histogram. Percentile mean and SD measures are calculated from the top 50%, 25%, and 10% of the histogram curve. Skewness is zero if the data are distributed symmetrically around the mean, negative if the data are more spread out on the left of the mean, and positive if the data are more spread out on the right of the mean. Kurtosis, a measure of how “peaked” the histogram is, equals zero if the histogram is Gaussian, is positive if the histogram has a sharper peak, and is negative if it has a flatter top.

Fig 2.

Low-grade glioma (grade II/IV) in left frontal lobe, T2-weighted (A) and contrast T1-weighted (B) images. The rCBV_max method uses 4 small ROIs targeted to foci of greatest perfusion on the rCBV map (C), with the maximal rCBV recorded from the subsequent perfusion curves (E). The signal intensity curves from each of the 5 ROIs are denoted as S1, S2, S3, S4, and S5, where S1 is the signal intensity curve for the ROI placed in normal brain and S2–S5 are the other ROIs placed in the tumoral tissue. These 5 signal intensity curves were obtained from a single section from the perfusion dataset. The rCBV histogram method uses a single ROI (D) that encompasses the maximal tumor diameter to generate the histogram curve (F), from which multiple metrics are derived.

Fig 3.

High-grade glioma, glioblastoma multiforme (grade IV/IV) in frontal lobes spanning the corpus callosum. T2-weighted (A) and contrast T1-weighted (B) images are shown along with rCBV_max map (C) with ROIs targeted to avoid areas of radiologic necrosis to determine perfusion curves (E). rCBV histogram map (D) and histogram curve (F) are derived from the maximal tumor diameter regardless of heterogeneity.