Perfusion Measurement in Brain Gliomas with Intravoxel Incoherent Motion MRI

Patient Demographics

The present study was approved by the local ethics committee at University of Lausanne. Patient consents were waived. From May 2011 to December 2012, our clinical glioma protocol included an IVIM sequence, which replaced the standard diffusion-weighted sequence. This provided apparent diffusion maps, which are part of the standard glioma evaluation, as well as perfusion-weighted maps, which allowed our exploratory work. Our clinical glioma protocol also included T1-weighted, T2-weighted, DSC perfusion, and T1-weighted postgadolinium sequences. Twenty-one consecutive patients (17 males, 4 females; mean age, 52.3 ± 21.3 years; age range, 2–84 years; Table 1) who had preoperative MR imaging examination, had no relevant treatment history at the time of imaging (such as radio-, chemo-, or antiangiogenic therapy), and had consecutive histopathologic diagnoses were included in the study. The tumor grading was based on World Health Organization criteria and yielded 16 high-grade and 5 low-grade gliomas. Because of the low number of low-grade gliomas found, a complementary study was performed and included 6 further low-grade gliomas diagnosed on radiologic criteria only (On-line Appendix).

Conventional MR Imaging

Conventional MR imaging, DSC, and IVIM were performed during the same procedure to allow direct comparison. The imaging was performed on 3T MR imaging scanners (Trio, Verio, or Skyra; Siemens, Erlangen, Germany) equipped with 32 multichannel receiver head coils. Before the examination, an 18- to 20-ga needle was inserted in either the right or the left antecubital vein. Afterward, T1-weighted sagittal and T2-weighted axial images were acquired.

IVIM MR Imaging

A Stejskal-Tanner diffusion-weighted spin-echo EPI pulse sequence¹⁵ was used, with multiple b-values (0, 10, 20, 40, 80, 110, 140, 170, 200, 300, 400, 500, 600, 700, 800, 900 s/mm²) in 3 orthogonal directions, and the corresponding trace was calculated. A single acquisition was obtained (no average). The images were orientated axially with a section thickness of 4 mm, an FOV of 297 × 297 mm², and a matrix size of 256 × 256, yielding an in-plane resolution of 1.2 × 1.2 mm². Parallel imaging, with an acceleration factor of 2 and a 75% partial Fourier encoding allowed TR/TE = 4000/99 ms. The receiver bandwidth was 1086 Hz/pixel, and fat was suppressed with a spectrally selective saturation routine. Total acquisition time was 3 minutes 7 seconds.

DSC MR Imaging

A gadolinium-based agent (gadoterate meglumine, Dotarem; Guerbet, Paris, France) was intravenously injected at a dose of 0.2 mL per kilogram of body weight and at a rate of 3 mL/s, followed by a 20-mL saline flush. Standard echo-planar images were consecutively acquired (TR = 1950 ms, TE = 43 ms, section thickness = 6 mm, FOV = 230 × 230, acquisition matrix = 128 × 128). No leakage correction was performed. CBV, MTT, and CBF maps were computed from the DSC MR imaging data by using the commercially available software, syngoMR (Siemens). On the section of interest, a region of interest was placed on an identified artery. We selected ≥4 voxels containing the best arterial input function curves, from which the average was built. We then set the time ranges defining baseline, gadolinium entry, and recovery, before the DSC perfusion maps were automatically calculated on the basis of a γ variate fitting of the time-concentration curve.

Regions of Interest

Brain gliomas have very heterogeneous structures and, accordingly, only the region with the highest malignancy defines the pathologic grade of the lesion. Therefore, a region of interest was manually placed on each tumor area in consensus by 2 experienced neuroradiologists (P.M. and P.H.) who were blinded to the histopathology, in the region of the tumor with maximal IVIM perfusion fraction and in the contralateral white matter, both on a single axial section. Cystic, hemorrhagic, or necrotic areas were avoided by using conventional pre- and postcontrast MR images. If a tumor was centered on the midline, a region of interest was chosen in the white matter of one of both hemispheres. Mean IVIM region-of-interest size was 322 ± 187 mm². The corresponding ROIs were then drawn on the DSC images, manually matching the images by using anatomic landmarks. The relative CBV was calculated by dividing the value of the region of interest in the tumor by the value of the contralateral normal-appearing white matter. The mean DSC region-of-interest size was 342 ± 151 mm².

We also produced histograms by placing a region of interest on all whole tumors on the IVIM b0 images, including all sections where the tumor was visible, and encompassing as much of the tumor area as possible but excluding cystic or necrotic areas, with the help of conventional pre- and postcontrast MR images. For the control histogram, a region of interest was placed in all patients to comprise the full white matter of an axial section of the contralateral hemisphere to the tumor.

All ROIs were placed so that they included as little CSF or as few large vessels as possible.

IVIM Image Processing

The standard IVIM 2-compartment diffusion model was assumed,⁷ with a “microvascular” and a “nonvascular” compartment, having, respectively, a pseudodiffusion coefficient and an apparent diffusion coefficient. The percentage of incoherent signal arising from the microvascular compartment f is called the perfusion fraction. To obtain the IVIM parameters, the IVIM signal equation was fitted in 2 steps, as previously described,²⁰ first for b > 200 s/mm² for the single parameter D, then for all b and all parameters, while keeping D constant. This fit was done on a voxel-by-voxel basis, by using the Levenberg-Marquardt algorithm²¹ implemented within Matlab (MathWorks, Natick, Massachusetts). This 2-step method increases robustness under biologic conditions and assumes that D* is significantly greater than D²² so that the influence of pseudodiffusion on signal decay can be neglected for b-values > 200 s/mm². Values under 0 for f, D, and D*, and values with f > 0.3 and D* > 0.05 mm²/s were considered not physiologic and were set to 0.²³ This step excluded poorly fitted voxels (< 1% of the voxels) and was necessary to ensure that a single artifactually very high value in a single voxel could not be responsible for the measured effect, while also increasing the contrast of the IVIM maps.

Statistical Analysis

Statistical analysis was performed with Excel (Microsoft, Redmond, Washington). A normal distribution of the data was assumed. Single-tailed, pair-wise Student t tests were calculated when data were compared with the contralateral region. Single-tailed, 2-sample, unequal-variance Student t tests were calculated between high- and low-grade tumor groups. Statistical significance was defined at P < .05. The Pearson r correlation coefficient between IVIM perfusion fraction f and DSC CBV was calculated.