Arvin Arani1, Christopher G. Schwarz1, Matthew C. Murphy1, Joshua D. Trzasko1, Jeffrey L. Gunter1, Matthew L. Senjem1, Heather J. Wiste1, Kiaran P. McGee1, Matthew A. Bernstein1, John Huston III1, and Clifford R. Jack Jr.1
1Mayo Clinic, Rochester, MN, United States
Synopsis
In magnetic resonance imaging (MRI) many factors can
contribute to non-tissue specific image intensity inhomogeneity. However, the
potential clinical impact or systematic biases of these effects have not been extensively
investigated across multiple MRI vendors and models for neuroimaging
applications. Specifically, left-right intensity comparisons are commonly used
by radiologists to verify/identify pathology. If significant systematic
left-right intensity asymmetries (LRIA) exist, it may lead to diagnostic
uncertainty and result in unnecessary imaging follow-up and patient burden. This
study shows that LRIA are common, system specific, systematic, can mimic
disease, create diagnostic uncertainty, and can impact multiple sequences
(T1-weighted and FLAIR).
Introduction
Localized
regions of inhomogeneity were observed at our institution on a Mayo Clinic
Study of Aging (MCSA) participant, on both T2 3D FLAIR and T1-weighted (T1-w) diagnostic
images (Figure 1). These images lead to suspicion of potential herpes
encephalitis and required unnecessary follow-up imaging. In magnetic resonance
imaging (MRI) many factors can contribute to non-tissue specific image
intensity inhomogeneity including static field inhomogeneity (Bo),
non-linear gradients(1), gradient induced eddy currents, bandwidth
filtering of data, and radio frequency transmission and reception inhomogeneity(2). Efforts
to correct MRI intensity inhomogeneity date back to 1986(3,4); however, the potential clinical impact or the
systematic biases that may occur on different MRI systems and models have not
been extensively investigated for neuroimaging applications. Specifically,
left-right intensity comparisons are commonly used by neuro-radiologists to
identify pathology. Therefore, if significant non-biological systematic
left-right intensities asymmetries (LRIA) exist, it may lead to diagnostic
uncertainty and unnecessary patient burden. The first objective of this
manuscript is to investigate if significant systematic LRIA exist for a range
of scanner models in study participants from the Alzheimer’s Disease
Neuroimaging Initiative (ADNI), where such variations are not expected to be
pathological. The second objective of this study is to investigate if LRIA can
introduce diagnostic uncertainty.Methods
T1-w and T2 FLAIR MRI were obtained
from the ADNI phases 2, GO, and 3 (www.adni-info.org). We used all available
pairs of 3T FLAIR MRI and 3T T1-weighted MRI. ADNI MRI acquisition protocols
have been previously published (5-7). Images were
downloaded, converted to NIfTI format, and 3D gradient inhomogeneity correction
was additionally applied for images where it was not already applied on-scanner
(1).
For atlas segmentation purposes, T1-w MR images were corrected for intensity
inhomogeneity and tissue-class probabilities were estimated by Unified
Segmentation (8) in SPM12 with tissue
priors and settings from the Mayo Clinic Adult Lifespan Template (MCALT; https://www.nitrc.org/projects/mcalt/)
(9). Atlas regions were
localized using Advanced Normalization Tools (ANTs) Symmetric Normalization (10) with the MCALT T1
template and ADIR122 atlas. To localize regions in FLAIR images, they were
rigid-registered to T1-w images in SPM12.
We additionally tested inhomogeneity
correction using the ADNI “N3m” preprocessing pipeline that was used to
preprocess T1-w images in ADNI 1-2 and refer to these as T1m images from hereon.
This pipeline uses the N3 algorithm for intensity homogeneity (11), followed by
additional correction using Unified Segmentation (8) in SPM5. To be
consistent within this work, we used the same regional parcellations from the
above SPM12/ANTs-based pipeline for estimating regional intensity values from
these SPM5-based N3m images.
Region mean intensity values for each
image volume were used to calculate the percent difference
(200*(left-right)/(left+right)) in each ROI. For each scanner model the percent
difference value for each exam was computed separately for the following image
types: 3D FLAIR, 2D FLAIR, T1-w, and T1m.
To
understand if asymmetry in left-right signal intensity impacted diagnostic
uncertainty we conducted a diagnostic survey. Two radiologists (CRJ, JH) were
shown thirty participants’ 3D FLAIR scans with hippocampal percent differences ranging
from -22% to 33% as determined from the ROI analysis. Each radiologist gave
separate scores (with a primary focus on the hippocampus) on the severity of
LRIA (1(no-asymmetry) – 5 (severe asymmetry)) and assessed whether they would
recommend clinical follow-up based on this LRIA.Results
The radiologists’ assessment of LRIA and follow-up
recommendations are shown in Figure 2. These results show that both clinicians
were uncertain or would recommend definite diagnostic follow-up in 45% of cases, with 70% of
these cases occurring when LRIA was above 10%. These images were all from a sample
where this type of pathology is not expected.
Percent difference in hippocampal intensities across
different scanner models used in ADNI is shown in Figure 3. Twenty-one scanner
models were used and the number of scans used in each box plot (n) has been
reported in brackets. The boxes represent the 1st quartile, median, and 3rd
quartiles, and the whiskers represent 1.5 standard deviations from the mean.
Data points that fell outside 1.5 standard deviations from the mean are plotted
(+). For 3 different scanner models the ROIs
with statistically significant LRIA, after Bonferroni correction, are shown for
3D FLAIR (Figure 4) and T1-w (Figure 5) images.
The yellow and blue overlays represent statistically significant left
side intensity hyper- and hypo-intensities, respectively.Discussion and Conclusions
This
study shows that LRIA can mimic disease and create uncertainty in diagnosis, resulting
in unnecessary patient burden. Statistically significant LRIA exist across
patients within specific scanner models for both T1-w and T2 FLAIR images.
These results also suggest that the polarity of the left-right asymmetry can
change, depending on the scanner model. The
post-processing techniques used to obtain T1m volumes were shown to be capable
of significantly improving LRIA (Figure 3), though they could potentially also remove
some true pathology. Hardware and software physics-aware on-scanner
inhomogeneity corrections, and lower tolerances in scanner calibrations, could
potentially reduce the impacts of this phenomenon. Until that is possible, it is important that
clinicians be aware of this phenomenon in order to reduce the risk of unnecessary
patient burden.Acknowledgements
This
work was supported by National Institutes of Health grants K12HD65987-11 and U19AG24904-15.References
1. Jovicich J, Czanner S, Greve D,
Haley E, van der Kouwe A, Gollub R, Kennedy D, Schmitt F, Brown G, MacFall J,
Fischl B, Dale A. Reliability in multi-site structural MRI studies: Effects of
gradient non-linearity correction on phantom and human data. NeuroImage
2006;30(2):436-443.
2. Vovk U, Pernus F, Likar B. A Review
of Methods for Correction of Intensity Inhomogeneity in MRI. IEEE transactions
on medical imaging 2007;26(3):405-421.
3. Haselgrove J, Prammer M. An
algorithm for compensation of surface-coil images for sensitivity of the
surface coil. Magnetic resonance imaging 1986;4(6):469-472.
4. McVeigh ER, Bronskill MJ, Henkelman
RM. Phase and sensitivity of receiver coils in magnetic resonance imaging.
Medical physics 1986;13(6):806-814.
5. Jack CR, Bernstein MA, Borowski BJ,
Gunter JL, Fox NC, Thompson PM, Schuff N, Krueger G, Killiany RJ, DeCarli CS,
Dale AM, Carmichael OW, Tosun D, Weiner MW. Update on the Magnetic Resonance
Imaging core of the Alzheimer's Disease Neuroimaging Initiative. Alzheimer's
& Dementia 2010;6(3):212-220.
6. Jack CR, Jr., Bernstein MA, Fox NC,
Thompson P, Alexander G, Harvey D, Borowski B, Britson PJ, J LW, Ward C, Dale
AM, Felmlee JP, Gunter JL, Hill DL, Killiany R, Schuff N, Fox-Bosetti S, Lin C,
Studholme C, DeCarli CS, Krueger G, Ward HA, Metzger GJ, Scott KT, Mallozzi R,
Blezek D, Levy J, Debbins JP, Fleisher AS, Albert M, Green R, Bartzokis G, Glover
G, Mugler J, Weiner MW. The Alzheimer's Disease Neuroimaging Initiative (ADNI):
MRI methods. Journal of magnetic resonance imaging : JMRI 2008;27(4):685-691.
7. Team TAMC. ADNI MRI SCANNER
PROTOCOLS. p WWW Document.
8. Ashburner J, Friston KJ. Unified
segmentation. NeuroImage 2005;26(3):839-851.
9. Schwarz CG, Gunter JL, Ward CP,
Vemuri P, Senjem ML, Wiste HJ, Petersen RC, Knopman DS, Jack Jr. CR. [P2–415]:
THE MAYO CLINIC ADULT LIFESPAN TEMPLATE: BETTER QUANTIFICATION ACROSS THE
LIFESPAN. Alzheimer's & Dementia 2017;13(7S_Part_16):P792-P792.
10. Avants BB, Epstein CL, Grossman M, Gee
JC. Symmetric diffeomorphic image registration with cross-correlation:
Evaluating automated labeling of elderly and neurodegenerative brain. Medical
Image Analysis 2008;12(1):26-41.
11. Sled
JG, Zijdenbos AP, Evans AC. A nonparametric method for automatic correction of
intensity nonuniformity in MRI data. IEEE transactions on medical imaging
1998;17(1):87-97.