Histogram analysis of MR imaging-derived cerebral blood volume maps: combined glioma grading and identification of low-grade oligodendroglial subtypes

Abstract

Background and purpose: Inclusion of oligodendroglial tumors may confound the utility of MR based glioma grading. Our aim was, first, to assess retrospectively whether a histogram-analysis method of MR perfusion images may both grade gliomas and differentiate between low-grade oligodendroglial tumors with or without loss of heterozygosity (LOH) on 1p/19q and, second, to assess retrospectively whether low-grade oligodendroglial subtypes can be identified in a population of patients with high-grade and low-grade astrocytic and oligodendroglial tumors.

Materials and methods: Fifty-two patients (23 women, 29 men; mean age, 52 years; range, 19-78 years) with histologically confirmed gliomas were imaged by using dynamic susceptibility contrast MR imaging at 1.5T. Relative cerebral blood volume (rCBV) maps were created, and 4 neuroradiologists defined the glioma volumes independently. Averaged over the 4 observers, a histogram-analysis method was used to assess the normalized histogram peak height of the glioma rCBV distributions.

Results: Of the 52 patients, 22 had oligodendroglial tumors. The histogram method was able to differentiate high-grade gliomas (HGGs) from low-grade gliomas (LGGs) (Mann-Whitney U test, P < .001) and to identify low-grade oligodendroglial subtypes (P = .009). The corresponding intraclass correlation coefficients were 0.902 and 0.801, respectively. The sensitivity and specificity in terms of differentiating low-grade oligodendroglial tumors without LOH on 1p/19q from the other tumors was 100% (6/6) and 91% (42/46), respectively.

Conclusion: With histology as a reference, our results suggest that histogram analysis of MR imaging-derived rCBV maps can differentiate HGGs from LGGs as well as low-grade oligodendroglial subtypes with high interobserver agreement. Also, the method was able to identify low-grade oligodendroglial tumors without LOH on 1p/19q in a population of patients with astrocytic and oligodendroglial tumors.

Fig 1.

A, Axial T2-weighted fast spin-echo (FSE) image (TR/TE, 4000/104 ms) of a patient (subject 3) with a low-grade oligoastrocytoma without LOH on 1p/19q. B, Axial postcontrast T1-weighted spin-echo (SE) image (TR/TE, 500/7.7 ms) of subject 3. C, Coregistered rCBV map of subject 3. D, Axial T2-weighted FSE image (TR/TE, 4000/104 ms) of a patient (subject 7) with a low-grade oligoastrocytoma with LOH on 1p/19q. E, Axial postcontrast T1-weighted SE image (TR/TE, 500/7.7 ms) of subject 7. F, Coregistered rCBV map of subject 7. Note the low rCBV values in the tumor area in image C compared with the tumor area in image F, typical of low-grade oligodendroglial tumors without LOH on 1p/19q. The corresponding normalized histogram signatures are shown in Fig 2.

Fig 2.

Resulting normalized histogram plots of the total distribution of rCBV values from the patients shown in Fig 1. The higher peak height of the low-grade oligoastrocytoma without LOH on p/19q, shown in a dotted line (subject 3, Fig 1A–C), indicates a more homogeneous rCBV distribution than the rCBV distribution of a low-grade oligoastrocytoma with LOH on 1p/19q, shown in a solid line (subject 7, Fig 1D–F).

Fig 3.

Mean histogram peak heights with SEs of the mean for the different glioma types investigated. The values are as follows: HGG = 0.052 (0.004), LGG = 0.094 (0.004), all high-grade oligodendroglial tumors (HGO) = 0.068 (0.010), low-grade oligodendroglial tumor with LOH on 1p/19q (LGO[−]) = 0.096 (0.004), low-grade oligodendroglial tumor without LOH on 1p/19q (LGO[+]) = 0.113 (0.002), and low-grade diffuse astrocytoma (LGA) = 0.086 (0.012). The oligodendroglial tumors include both oligodendrogliomas and oligoastrocytomas.

Fig 4.

A scatterplot showing histogram peak heights for the 52 patients included in our study. The histogram method is able to significantly differentiate between HGGs (▴) and LGGs (+) (P < .001), with an ICC of 0.902.

Fig 5.

Histogram peak heights for the 52 patients investigated in our study (23 LGGs and 29 HGGs). Each histogram peak height is a mean value across the 4 observers. With a cutoff value of 0.107, the sensitivity and specificity when distinguishing the 6 patients with low-grade oligodendroglial tumors without LOH on 1p/19q (red bars) from the other 46 patients (blue bars) are 100% (6/6) and 91% (42/46), respectively.

Fig 6.

ROC curve for the histogram method when distinguishing patients with low-grade oligodendroglial tumors without LOH on 1p/19q (n = 6) from the other gliomas (n = 46). The area (±SE) under the ROC curve is A_z = 0.931 ± .036.