SUBJECTS
We screened subjects with intra-axial brain mass suggestive of high-grade glioma, who were referred to the radiology department of the Hospital of University of Pennsylvania from 2016 to 2019. Twenty-one adult patients who underwent tumor resection and whose tissue specimens were consistent with the histopathological diagnosis of GBM were enrolled in the study.
This study was approved by the Institutional Review Board of the University of Pennsylvania, and all research was performed in accordance with relevant guidelines and regulations. Informed consent has been obtained from patients to include the additional research MRI sequence and histopathological studies as described in the following sections.
Image Acquisition
Imaging was performed on a 3 T scanner (Trio; Siemens) equipped with a 12-channel head coil. Single echo gradient echo MRI was obtained in the axial orientation using the following imaging parameters: in-plane spatial resolution = 0.86 × 0.86 mm2, slice thickness = 3 mm, matrix =256 × 256 × 24, flip angle (FA) = 20, TE/TR = 18/55 msec, and bandwidth = 444 Hz/pixel. Additional imaging included post-contrast T1 magnetization-prepared rapid gradient-echo (MPRAGE) (in-plane spatial resolution = 0.977 × 0.977 mm2, slice thickness = 1 mm, matrix =256 × 192 × 192, TE/TR/TI = 3.11/1750/950 msec, FA = 15, and bandwidth = 150 Hz/pixel) and fluid-attenuated inversion recovery imaging (FLAIR) (in-plane spatial resolution = 0.938 × 0.938 mm2, slice thickness = 3 mm, matrix =256 × 192 × 60, TE/TR/TI = 141/9420/2500 msec, FA = 170, and bandwidth = 287 Hz/pixel) after intravenous administration of gadolinium-containing contrast material (MultiHance, Bracco, Princeton, NJ or Dotarem, Guerbet); 0.1 mmol/kg, double dose. DSC-MR imaging was performed by a gradient-echo echo-planar (GE-EPI) imaging sequence during a second 0.1-mmol/kg bolus of contrast with the following parameters: TR/TE = 2000/45 ms, FOV = 22 × 22 cm2, resolution = 1.72 × 1.72 × 3 mm3, 20 sections.
For ex vivo imaging, we selected three subjects whose specimens had the least hemorrhage in order to minimize artifact in susceptibility maps on ultra-high field MRI. The formalin-fixed paraffin-embedded tissue samples for these subjects were imaged on a Bruker 9.4 Tesla 8.9 cm vertical bore MR as explained in detail in the supplementary material (see supplementary figure 1). T2 TurboRARE, and 3D multi-echo GRE scans were acquired at 60°C to ensure paraffin melt (See supplementary material and methods).
Histology.
A board-certified neuropathologist (M.P.N.), who was blinded to MRI susceptibility measurements, assessed the specimens from the initial surgical resections. Nineteen subjects had enough tissue for histological studies. Samples were formalin-fixed and paraffin-embedded, and the block that best represented the submitted tissue (often greater than 50% of the entire tissue) was chosen for additional staining.
Each specimen was stained with hematoxylin and eosin, as well as with CD68, CD86, CD206, and L-Ferritin. The unstained slides underwent heat-induced epitope retrieval in citrate buffer pH 6.0 (Leica Microsystems) for 20 min. We then performed immunohistochemical staining on the Bond 111 Autostainer with the hematoxylin counterstain and DAB chromogen.
For all subjects, histology slides were digitized at the Children Hospital of Philadelphia pathology core facility at 20 X using an Aperio CS-O slide scanner (Leica Biosystems, Buffalo Grove, IL). These scanned images were preprocessed with QuPath software (22), using an automatic module to adjust the IHC stain vectors and remove unexpected colors due to artifacts. Areas of artifact (e.g. tissue folds, edge artifacts and scanning artifacts) were removed from the analysis through manual tissue segmentation using the QuPath annotations tool (22). For each case, the positivity percentage of the IHC images (i.e. CD68, CD86, CD206, and FL chain) was then quantified for the whole slide based on a threshold applied to the 3,3' Diaminobenzidine (DAB) signal in the cytoplasm. Refer to the supplementary material table-1 for the parameters used in positive cell detection. To address variation observed in staining intensity, for each IHC image, positive cell quantification was first performed using an initial threshold equal to mean + 2SD of DAB optic density of all cells. To minimize the error rate in positive cell detection, this threshold was then modified by a neuropathologist (M.P.N) accordingly.
Radiologic image analysis:
All sequences were registered to post-contrast T1. using the FSL MRI toolkit. Regions of abnormal contrast enhancement, necrosis, and non-enhancing FLAIR signal intensity were segmented using a semi-automated segmentation tool (ITK- SNAP)(23) followed by manual editing by two board-certified neuroradiologists (S.A.N. and J.B.W.). Areas of susceptibility artifact on source DSC (dynamic susceptibility contrast) and areas with high signal intensity on pre-contrast T1 source were excluded from these segmentations as areas of hemorrhage.
QSM images were reconstructed from single echo gradient-echo MRI magnitude and phase data using the morphology-enabled dipole inversion (MEDI) algorithm(24), implemented in MATLAB (MathWorks, Natick, Massachusetts). A brain mask was obtained from the magnitude image using the FSL brain extraction tool (25). Subsequently, phase unwrapping was performed using Laplacian unwrapping (26), transmit phase was removed by fitting and subtracting a fourth-order 3D polynomial(27), and background field removal was performed using the regularization-enabled sophisticated harmonic artifact reduction for phase data (RESHARP) algorithm(28), prior to susceptibility inversion with MEDI using the default regularization parameters.
A threshold of 0.055 was applied to QSM values to exclude areas of artifact and venous structures. The thresholded QSM maps were reviewed and confirmed by a board-certified neuroradiologist (S.A.N.), who was blind to the immunohistochemistry results. The above-mentioned segmentations were applied to the QSM images and the voxel count and the average susceptibility of the segmented regions were recorded for each subject.
The T2* maps for the two ex-vivo samples were generated by a pixel-wise monoexponential fitting of the multi-echo GRE images using MATLAB (29).
Pathology-radiology region of interest (ROI) analysis:
For the cases with ex vivo MRI, we undertook an ROI approach for direct correlation analysis between the corresponding ROIs on IHC and MRI images. First, areas of highly dense tumor were determined by a tissue classification model (1µm/pixel; random trees; QuPath 2.0.1)(22) that was trained by a neuropathologist (M.P.N) on H&E images (figure 2). Subsequently, 16 ROIs (0.25 mm2 each) were randomly placed in areas of tumor. The number of ROIs per slide was proportional to the total area of tumor dense regions on each slide. For each ROI on the corresponding IHC images, the total number of positive cells was counted as described above. To transfer ROIs to the corresponding coordinates on the T2* maps, IHC images were co-registered with MRI images using HistoloZee software (see Figure 3). The mean T2* values (19.5 ± 11.7)(23) were then used to calculate the effective transverse relaxivity (R2*= 1/T2*) for each ROI.
Statistical Analysis
We performed Shapiro -Wilk test to assess normality of data. For the correlation analyses between MRI metrics and histology, we used Pearson’s R for normally distributed data and Spearman’s Rho for non-normally distributed data. Mean and standard deviation was reported for descriptive statistics. P values below 0.05 were considered statistically significant. Statistical analysis was conducted in the R statistical environment (version 3.6.1; http://www.r-project.org/).