Histogram analysis of tensor-valued diffusion MRI in meningiomas: Relation to consistency, histological grade and type

Abstract

Background: Preoperative radiological assessment of meningioma characteristics is of value for pre- and post-operative patient management, counselling, and surgical approach.

Purpose: To investigate whether tensor-valued diffusion MRI can add to the preoperative prediction of meningioma consistency, grade and type.

Materials and methods: 30 patients with intracranial meningiomas (22 WHO grade I, 8 WHO grade II) underwent MRI prior to surgery. Diffusion MRI was performed with linear and spherical b-tensors with b-values up to 2000 s/mm². The data were used to estimate mean diffusivity (MD), fractional anisotropy (FA), mean kurtosis (MK) and its components-the anisotropic and isotropic kurtoses (MK_A and MK_I). Meningioma consistency was estimated for 16 patients during resection based on ultrasonic aspiration intensity, ease of resection with instrumentation or suction. Grade and type were determined by histopathological analysis. The relation between consistency, grade and type and dMRI parameters was analyzed inside the tumor ("whole-tumor") and within brain tissue in the immediate periphery outside the tumor ("rim") by histogram analysis.

Results: Lower 10th percentiles of MK and MK_A in the whole-tumor were associated with firm consistency compared with pooled soft and variable consistency (n = 7 vs 9; U test, p = 0.02 for MK_{A 10} and p = 0.04 for MK₁₀) and lower 10th percentile of MD with variable against soft and firm (n = 5 vs 11; U test, p = 0.02). Higher standard deviation of MK_I in the rim was associated with lower grade (n = 22 vs 8; U test, p = 0.04) and in the MK_I maps we observed elevated rim-like structure that could be associated with grade. Higher median MK_A and lower median MK_I distinguished psammomatous type from other pooled meningioma types (n = 5 vs 25; U test; p = 0.03 for MK_{A 50} and p = 0.03 and p = 0.04 for MK_{I 50}).

Conclusion: Parameters from tensor-valued dMRI can facilitate prediction of consistency, grade and type.

Keywords: Consistency; Diffusion MRI; Meningioma; Tensor-valued diffusion encoding; Tumor grade; Type.

Fig. 1

Contrast overview (panel A), flow diagram (B) and ROI definition (C). Panel A shows an example of a WHO grade I fibroblastic meningioma with variable consistency. Note that FA is high in the tumor periphery and central part but lower medially from the central part. That indicates that FA and MK_A reflect two different aspects - MK_A maps the microscopic diffusion anisotropy whereas FA shows the macroscopic diffusion anisotropy, which is lower due to low orientation coherence (Szczepankiewicz et al., 2016). Panel B shows the flow diagram of the study where the two patient populations are characterized in Table 1, respectively. The consistency analysis was performed on 16 subjects while tumor grade and tumor type analysis on 30 subjects. Panel C defines two region-of-interests (ROIs) used in the study – a “whole-tumor ROI” characterizing inner parts of the tumor and “rim ROI” characterizing the reaction in the brain tissue surrounding the tumor.

Fig. 2

Histograms of dMRI parameters. Panel A shows distributions of parameter-values within the ROIs. Part 1 shows MD, MK and MK_A distributions in the whole-tumor ROI among soft, variable, and firm consistency (distribution differences indicated by yellow arrows). Part 2 shows distributions of MK_A, MK_I and MK in psammomatous and other meningioma types in the whole-tumor ROI. Part 3 shows the MK_I distribution in grade I and II meningiomas within the rim ROI. The histograms suggest that it may be valuable to consider different distribution characteristics—their percentiles or standard deviation—shown by yellow arrows. Panel B shows effect sizes for different distribution characteristics (10th, 25th, 50th, 75th, 90th and standard deviation) as measured by Cohen’s d (defined according to Eq. (2)) averaged across all dMRI parameters (MD, FA, MK, MK_A and MK_I and S₀). For consistency, grade and type the highest effect size is found for the 10th, 10th, and 50th percentile, respectively. For the case of grade, however, only standard deviation within MK_I was significantly different between the grades. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3

Consistency estimation. Panel A shows dMRI parameters (10th percentile within whole-tumor ROI) versus meningioma consistency. In total we included 16 patients (with meningiomas of 7 firm, 5 variable and 4 soft). Based on MK_{A 10} and MK₁₀ the firm consistency is significantly different from pooled soft and variable consistency (U test, p = 0.04 for MK₁₀, p = 0.02 for MK_{A 10}). Based on MD₁₀, the variable consistency can be distinguished from soft and firm one (U test, p = 0.02). Significant parameters are marked with an asterisk (*), significant distributions with yellow arrows. Panel B shows two examples from panel A where MK₁₀ may be useful on the individual level (marked by yellow markers in panel A). The top row shows a firm tumor that is considerably darker on the MK map in comparison to the non-firm tumor in the bottom row (yellow arrows). The images are scaled according to scale bars from Fig. 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4

Grade estimation. Panel A shows a grade I versus grade II comparison of the standard deviation in the rim ROI for the different dMRI parameters. Standard deviation within the rim ROI of MK_I of grade I was significantly higher than that of grade II meningioma (n = 22 vs 8; U test; p = 0.04). All tumors were classified according to WHO 2016 classification (Louis et al., 2016).

Fig. 5

MK_I-rim as a radiological feature. We have observed a presence of elevated MK_I values in a rim-like structure that partially circumscribes the tumor (panel A) or its absence (panel B). Yellow arrows mark the MK_I-rim in panel A or tumor region in panel B, respectively. The MK_I-rim is found in the brain tissue surrounding the T1w + Gd enhancing tumor lesion and it was preferentially present in high grade meningiomas. All cases can be found in the supplementary material. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 6

Type estimation. Panel A shows 50th percentile (median) of dMRI parameters versus meningioma type, where 1 = Fibroblastic, 2 = Meningothelial, 3 = Transitional, 4 = Clear-cell, 5 = Microcystic/Angiomatous, 6 = Chordoid, 7 = Psammomatous. The microcystic/angiomatous type (#5) is significantly different from the other types in S₀, MD, FA, MK, MK_A, however, only two cases were included. The psammomatous type (#7) is significantly different from the others in MK_{A 50} and MK_{I 50} but not in MK₅₀ (n = 5 vs 24; U test; p = 0.03 for MK_{A 50} and p = 0.03 and p = 0.04 for MK_{I 50}). Panel B shows three examples from panel A (marked by yellow markers). The one in the top row shows a psammomatous type with high MK_{A 50}. The middle row shows a case of non-psammomatous (fibroblastic) type with the highest MK_{A 50}. The psammomatous type is somewhat brighter in MK_A map than the highest non-psammomatous type (yellow arrows). The bottom row shows that an example of MD being generally high in a microcystic/angiomatous meningioma (yellow arrows). The images are scaled according to scale bars from Fig. 1. All tumors were classified according to WHO 2016 classification (Louis et al., 2016). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 7

Receiver-operating characteristic (ROC) curves for the 3 cases from Fig. 1. Panel A shows ROC curves for predicting firm consistency (n = 7) from the pooled soft and variable (n = 9) based on MK₁₀ (gray line) and MK_{A 10} (black line) in the whole tumor-ROI. Based on MK₁₀ the AUC is 0.83 with confidence interval [0.43; 1.00] with optimal cut-point value (gray dot) with specificity 100 % and sensitivity 71 %. Based on MK_{A 10} the AUC is 0.84 with confidence interval [0.52; 0.98] with optimal cut-point value (black dot) with specificity 78 % and sensitivity 86 %. Panel B shows ROC curve for predicting grade II (n = 8) from grade I (n = 22) based on MK_{I std} in the rim-ROI. The AUC is 0.65 with confidence intervals [0.42; 0.84]. Optimal cut-point value (gray dot) yields specificity of 77 % and sensitivity 50 %. Finally, panel C shows ROC curve for prediction of psammomatous type based on MK₅₀ and MK_{A 50}. The AUC for MK₅₀ is 0.63 with confidence interval [0.18; 1.00] with optimal cut-point value with specificity of 100 % and sensitivity 40 % (gray dot) and for MK_{A 50} 0.81 with confidence intervals [0.07; 1.00] with optimal cut-point value with specificity 96 % and sensitivity 80 %. Large confidence intervals (not shown graphically in the figure) in all cases highlight limitations of the small dataset and thus limit the interpretability of the AUC value.