A new method for quantification and 3D visualization of brain tumor adhesion using slip interface imaging in patients with meningiomas

Abstract

Objectives: To develop an objective quantitative method to characterize and visualize meningioma-brain adhesion using MR elastography (MRE)-based slip interface imaging (SII).

Methods: This retrospective study included 47 meningiomas (training dataset: n = 35; testing dataset: n = 12) with MRE/SII examinations. Normalized octahedral shear strain (NOSS) values were calculated from the acquired MRE displacement data. The change in NOSS at the tumor boundary (ΔNOSSbdy) was computed, from which a 3D ΔNOSSbdy map of the tumor surface was created and the probability distribution of ΔNOSSbdy over the entire tumor surface was calculated. Statistical features were calculated from the probability histogram. After eliminating highly correlated features, the capability of the remaining feature for tumor adhesion classification was assessed using a one-way ANOVA and ROC analysis.

Results: The magnitude and location of the tumor adhesion can be visualized by the reconstructed 3D ΔNOSSbdy surface map. The entropy of the ΔNOSSbdy histogram was significantly different between adherent tumors and partially/completely non-adherent tumors in both the training (AUC: 0.971) and testing datasets (AUC: 0.900). Based on the cutoff values obtained from the training set, the ΔNOSSbdy entropy in the testing dataset yielded an accuracy of 0.83 for distinguishing adherent versus partially/non-adherent tumors, and 0.67 for distinguishing non-adherent versus completely/partially adherent tumors.

Conclusions: SII-derived ΔNOSSbdy values are useful for quantification and classification of meningioma-brain adhesion. The reconstructed 3D ΔNOSSbdy surface map presents the state and location of tumor adhesion in a "clinician-friendly" manner, and can identify meningiomas with a high risk of adhesion to adjacent brain parenchyma.

Key points: • MR elastography (MRE)-based slip interface imaging shows promise as an objective tool to preoperatively discriminate meningiomas with a high risk of intraoperative adhesion. • Measurement of the change of shear strain at meningioma boundaries can provide quantitative metrics depicting the state of adhesion at the tumor-brain interface. • The surface map of tumor adhesion shows promise in assisting precise adhesion localization, using a comprehensible, "clinician-friendly" 3D visualization.

Keywords: Elasticity imaging techniques; Magnetic resonance; Meningioma; Tissue adhesions.

Conflict of interest statement

Conflict of interest The Mayo Clinic and the authors of this manuscript have intellectual property and a financial interest related to this research.

This research has been reviewed by the Mayo Clinic Conflict of Interest Review Board and is being conducted in compliance with Mayo Clinic Conflict of Interest policies.

© 2021. European Society of Radiology.

Figures

Fig. 1

(a) Brain MR elastography (MRE) setup. (b–g) The diagram of the ΔNOSSbdy calculation. (b) An example slice of T1W image with a manually delineated tumor margin. (c) The tumor shell after dilation, erosion, and exclusion of non-interface regions. (d) Black lines orthogonal to the tumor surface placed at contour points. For illustration purposes, only half of the orthogonal lines are shown. (e) The surface mesh is generated by MATLAB from the tumor volume and smoothed by a box filter with the kernel size of 3 × 3 × 3. (f) The 3D surface map with ΔNOSSbdy values projected onto the 3D surface. The gray color indicates a non-interface region that was not included in the calculation. (g) A 2D Hammer projection of the ΔNOSSbdy surface map unfolded along the z-direction to facilitate the 3D data visualization

Fig. 2

Flowchart of patient selection

Fig. 3

The probability histograms of ΔNOSSbdy values generated from the tumor shells for representative complete, partial, and non-adherent tumors. The histogram fitting curve of the adherent tumor is high and sharp, while the histogram fitting curve of the non-adherent tumor is wide and flat

Fig. 4

Group comparison of (a) entropy and (b) mean of the probability histogram of ΔNOSSbdy values among completely, partially, and non-adherent meningiomas in the training cohort. *p was tested with one-way ANOVA with post hoc Steel-Dwass multiple comparisons

Fig. 5

Receiver operating characteristic (ROC) curves for the entropy of the probability histogram of the ΔNOSSbdy values in the (a) training and (b) testing cohorts. (c, d) The ΔNOSSbdy entropy for each patient sorted by the classification of tumor adhesion in the training and testing cohorts. (c) Complete adhesion versus partial/no adhesion. (d) No adhesion versus partial/complete adhesion. The dotted lines represent the best cutoff values for distinguishing tumor adhesion

Fig. 6

Illustrative cases. (a) A 46-year-old female. The surgeons described the brain-tumor interface as extremely poor, and there was sub-pial invasion all around this tumor. Although the NOSS map shows a faint line of slightly increased NOSS value at the tumor periphery (arrow), both the 3D surface map of the ΔNOSSbdy and the 2D Hammer map indicate that the majority of the tumor surface was adhesive. (b) A 52-year-old female. It was noted at the surgery that this tumor had no adhesion on the medial side adjacent to the falx, but was adherent to the lateral side of the brain, which correlates well with the ΔNOSSbdy surface map. The NOSS map shows a faint hyper-intensity NOSS contour. (c) A 51-year-old female. This was a non-adherent tumor. Both the NOSS map and 3D surface map of the ΔNOSSbdy agreed well with the surgical evaluation. Rotating 3D animation of the tumor surface maps are shown in the electronic supplementary material