Ashley M Stokes1, Jack T Skinner2, Laura C Bell1, Adrienne N Dula3, Thomas E Yankeelov3, and C. Chad Quarles1
1Translational Bioimaging Group, Barrow Neurological Institute, Phoenix, AZ, United States, 2Imaging Programs, National Comprehensive Cancer Network (NCCN), Philadelphia, PA, United States, 3University of Texas - Austin
Synopsis
The purpose of this study
is to investigate the influence of post-processing method on the
reproducibility of brain diffusion metrics, including apparent diffusion
coefficients (ADCs) and intra-voxel incoherent motion (IVIM) parameters, in
healthy controls and to apply these results in a cohort of brain tumor patients
undergoing treatment. ADC was highly reproducible for all methods. The IVIM diffusion
and perfusion fraction showed the highest reproducibility using constrained
fitting, while IVIM pseudo-diffusion showed limited reproducibility. By establishing
limits of repeatability for ADC and IVIM metrics, these methods can be applied in
neuropathology to determine significant changes related to treatment effects.
Introduction
Diffusion-weighted imaging
(DWI) is a commonly employed biomarker of cellular status in in neuro-oncology.
For example, changes in the apparent [water] diffusion coefficient (ADCs)
reflect treatment-induced tumor cellular changes.1 However, a
critical obstacle for the development of diffusion-based biomarkers comes in
differentiating between a true treatment effect and confounding measurement
inaccuracies.2 Using a standard
post-processing method, both ADC and the intra-voxel incoherent motion (IVIM)
diffusion coefficient D were shown to
be highly reproducible across scanners, while the IVIM perfusion fraction fp showed moderate
reproducibility.3 As the choice of
post-processing method also affects the reliable derivation of diffusion
metrics,4 the purpose of
this study is (i) to investigate the influence of post-processing method on the
reproducibility of brain diffusion metrics in healthy controls and (ii) to
apply these results in a cohort of brain tumor patients undergoing treatment.Methods
Data were acquired at 3T
(Achieva, Philips Healthcare) in healthy controls (n = 9) and high-grade glioma
patients (n = 6) at three time points (time 0, 1 week, and 4 weeks). The
patient data corresponded with before and 2 time points after Bevacizumab
treatment. Images were acquired with a single-shot spin-echo EPI sequence with
TR = 6 s, TE = 50 ms, SENSE parallel imaging (2 in AP direction), voxel size = 2.5
mm isotropic, 40 slices, with three orthogonal diffusion-encoding
directions and 8 b-values (b = 0, 25, 50, 75, 100, 200, 500, 1000
s/mm2). All data were registered to the T1-weighted image
obtained at the 2nd time point using the affine (12 DOF)
registration algorithm FLIRT (FSL, FMRIB Centre, Univ. of Oxford). The
segmentation algorithm FAST (FSL) was used to segment white matter (WM) and
gray matter (GM) regions of interest (ROIs). The diffusion data were normalized
to the b = 0 data (S0) and then fit to
mono-exponential and bi-exponential equations to obtain ADC and IVIM
parameters. For ADC, four post-processing methods were compared, while three
post-processing methods were compared for IVIM (equations and methods are
listed in Table 1). Repeatability was assessed using previously published
methods.5Results / Discussion
Figure 1 (a-d) shows DWI
data and Method 1 fits for ADC (a,b) and IVIM (c,d) parameters in healthy
control WM (a,c) and GM (b,d) ROIs over three time points. Signal from each b-value
is consistent between time-points, yielding consistent ADC and IVIM-D
maps (Figure 1, e). Visually, IVIM-D*
did not appear reproducible across time points, while IVIM-fp appeared more consistent. Boxplots (Figure 2) for all
patients and time points demonstrate the influence of post-processing methods
on ADC and IVIM metrics. Across post-processing methods, the highest
coefficient of variation (CV) for ADC was observed for Method 1 (1.99%), though
not statistically different from Method 2-4 CV (1.86% for all). It should also
be noted that the ADC values for Method 1 and 4 are higher due to increased
sensitivity to perfusion effects at low b-values.6 For IVIM, Method 1
yielded the highest variability (CV = 5.5%, 8.1%, and 13% for IVIM-D, IVIM-fp, and IVIM-D*,
respectively). IVIM-fp had
the lowest CV (4.6%) for Method 2. Overall, IVIM-D* was not consistently fit by any method. Figure 3 shows
Bland-Altman mean-difference plots between T0 and T1 (1-week) in control WM, control GM,
and tumor ROIs using Method 1 ADC and Method 2 IVIM parameters. The
repeatability limits were less than ±0.1x10-3 mm2/s for ADC and IVIM-D and less than ±2% for IVIM-fp. IVIM-D*
showed wider limits of repeatability (< ±4x10-3 mm2/s). Using these limits applied to the
patient cohort, several individual tumors showed 1-week tumor diffusion changes
outside of the repeatability limits. Statistical analysis using repeated
measures analysis of variance is ongoing. Conclusions
We have demonstrated the
influence of post-processing methods on the derived diffusion metrics ADC,
IVIM-D, IVIM-fp, and IVIM-D*.
All four tested methods for ADC showed similar variance, though methods that
include data from b = 0 s/mm2
are sensitive to perfusion effects that may affect the resulting ADC values.
Simultaneous fitting for all IVIM parameters is not recommended due to the high
variability in the results parameters. IVIM-fp
can be reproducibly measured using either constrained fitting method, but IVIM-D* is not reproducible. This study has
permitted us to establish expected levels of variability in ADC and IVIM
metrics over time, and variations outside of these limits may indicate altered
pathology related to tumor growth or treatment effects. Acknowledgements
This work was supported by
NIH/NCI 1R01CA158079, NIH/NCI U01 CA142565, NIH/NCATS KL2 TR 000446, and
NIH/NCATS RR024975.References
1. Chenevert
TL, Stegman LD, Taylor JMG, et al. Diffusion Magnetic Resonance Imaging: an
Early Surrogate Marker of Therapeutic Efficacy in Brain Tumors. Journal of the
National Cancer Institute 2000;92(24):2029-2036.
2. Padhani
AR, Liu G, Mu-Koh D, et al. Diffusion-Weighted Magnetic Resonance Imaging as a
Cancer Biomarker: Consensus and Recommendations. Neoplasia 2009;11(2):102-125.
3. Grech-Sollars
M, Hales PW, Miyazaki K, et al. Multi-centre reproducibility of diffusion MRI
parameters for clinical sequences in the brain. Nmr Biomed 2015;28(4):468-485.
4. Meeus EM,
Novak J, Withey SB, Zarinabad N, Dehghani H, Peet AC. Evaluation of intravoxel
incoherent motion fitting methods in low-perfused tissue. Journal of Magnetic
Resonance Imaging 2016.
5. Galbraith
SM, Lodge MA, Taylor NJ, et al. Reproducibility of dynamic contrast-enhanced
MRI in human muscle and tumours: comparison of quantitative and
semi-quantitative analysis. Nmr Biomed 2002;15(2):132-142.
6. Cohen
AD, LaViolette PS, Prah M, et al. Effects of perfusion on diffusion changes in
human brain tumors. Journal of magnetic resonance imaging : JMRI
2013;38(4):868-875.