Benjamin Charles Musall1, Yanyu Yang2, Arash Kamali1, John A Lincoln3, Vi Ly2, Xi Luo2, Ponnada A Narayana1, Refaat E Gabr1, and Khader M Hasan1
1Diagnostic and Interventional Imaging, UTHealth McGovern Medical School, Houston, TX, United States, 2Department of Biostatistics and Data Science, University of Texas School of Public Health, Houston, TX, United States, 3Department of Neurology, UTHealth McGovern Medical School, Houston, TX, United States
Synopsis
Keywords: Multiple Sclerosis, Quantitative Imaging
Quantitative
T2 of diffusely-abnormal white matter (DAWM), assessed at baseline in a
population of 800 patients with relapsing-remitting MS with U-Net segmentation
and T2 mapping, is distinct and intermediate to T2-hyperintense focal lesions
and normal-appearing white matter (NAWM).
INTRODUCTION
MRI-based assessment of multiple sclerosis (MS) lesions show
only moderate correlation with clinical symptoms1,2. In addition to
focal lesions, diffuse regions of white matter with signal intensities intermediate
to normal-appearing white matter (NAWM) and focal lesions are seen on T2-weighted
and proton-density MRI. This diffusely-abnormal white matter (DAWM) is seen in more
than 25% of relapsing-remitting MS patients3. DAWM pathology is not
yet understood, and may be distinct from that of focal lesions3.
Expert segmentation of DAWM is time-intensive and difficult,
and automated methods are still under development3. Some preliminary
studies using quantitative MRI have shown DAWM to have tissue properties
distinct to NAWM and lesions3. Quantitative DAWM metrics may be
useful for development of automated assessments and are also of interest as
possible correlates with disability measures and disease progression.
In this report, we applied a U-Net model4 for segmentation
of normal-appearing gray matter (NAGM), NAWM, DAWM, and T2 lesions on baseline
scans of a large patient population with relapsing-remitting MS. The
segmentations were used to extract quantitative T2 measurements for comparison
between the tissues.METHODS
The baseline scans of 800 relapsing-remitting MS patients from
the multi-center CombiRx trial (NCT00211887) were analyzed. 223/800 (24%) of
patients were men. Patient age ranged from 18 to 61 years (median age: 37). Scans
were acquired using both 1.5 T and 3.0 T (73% at 1.5T) MRI scanners from
multiple vendors. Patient scans included fluid-attenuated inversion recovery (FLAIR),
pre-contrast T1-weighted, and dual spin-echo sequences. All series underwent
skull stripping and bias field correction5.
The dual spin-echo data were fit voxel-wise using a linear least-squares
solution of T2 exponential decay to estimate quantitative T2 maps. The TR and
TE of the dual echo scan were not fixed between sites, resulting in variable
parameter settings. Ranges of the TR and TE were as follows: TRs from 5320 to
7000 ms, early echo = TE1 from 7.2 to 16 ms, and late echo =TE2 from 85 to 115
ms.
A U-Net model6, trained by weak supervision and
refined on a small number of expert segmentations4, was used for
segmentation of the brain into NAGM, NAWM, cerebrospinal fluid (CSF), DAWM, and
T2 lesions (Figure 1).
Tissue volumes were calculated from U-Net segmentations. The
percentage of the intracranial volume occupied by DAWM was calculated. T2
measurements were extracted from T2 maps using tissue segmentations. T2 tissue
measurements were compared on a patient-wise basis using Wilcoxon Rank-Sum test.
A visual abstract depicting these methods can be seen in Figure 2.RESULTS
As segmented by the U-Net model, 81% of patients had >1 cm3
of DAWM, while 25% of patients has DAWM volumes of at least 6.6 cm3.
DAWM volumes ranged up to 32 cm3 (average: 4.8 ± 4.7 cm3),
while T2 lesion volumes ranged from 0.1 to 67.6 cm3 (average: 11.5 ± 12.0 cm3).
The percentage of the intracranial volume occupied by DAWM ranged from 0% to
2.6%. (average: 0.35% ± 0.34%).
Patient-wise comparisons (Figure 3) showed DAWM T2 to be
significantly higher than NAWM (p < 0.001) and significantly lower than T2
lesions (p < 0.001). A scatter plot showing average T2 values for tissues
across patients are shown in Figure 4.DISCUSSION
In a large population of relapsing-remitting MS patients, DAWM
was seen to have distinct T2 measurements in comparison to NAWM and T2 lesions.
This parallels the finding of a prior study by Papadaki et al., who found DAWM
in 27/37 (73%) relapsing-remitting MS patients with T2 values distinct to NAWM
and T2 lesions7. However; direct comparison of T2 values with this
study is not possible due to differences in DAWM segmentation technique and in
T2 acquisition, modeling and mapping techniques.
The heterogeneity of MRI hardware and
acquisition protocols in the scan dataset strengthens the finding of distinct
T2 in DAWM on a patient-wise basis, though this heterogeneity also introduces variability
in the T2 mapping. In addition, T2 values are known to vary by location in the
brain and in relation to patient demographics such as age and sex. Inter-patient
comparisons of T2 measurements and correlation with clinical endpoints may require
an accounting of these related factors.CONCLUSION
Our
preliminary results show that quantitative T2 measurements in DAWM are distinct
from T2 lesions and NAWM. Quantitative T2 may be useful for assessment of DAWM
in relapsing-remitting MS patients.Acknowledgements
No acknowledgement found.References
- Trapp B, Peterson J, Ransohoff, et al. Axonal
transection in the lesions of multiple sclerosis. N Engl J Med.
1998;338(5):278-85.
- Mollison D, Sellar R, Bastin M, et al. The
clinico-radiological paradox of cognitive function and MRI burden of white
matter lesions in people with multiple sclerosis: A systematic review and
meta-analysis. PLoS One. 2017;12(5):e0177727.
- Cairns J, Vavasour I, Traboulsee A, et al.
Diffusely Abnormal White Matter in Multiple Sclerosis. J Neuroimaging.
2022;32(1)5-16.
- Musall B, Kamali A, Lincoln J, et al. Weakly
Supervised Convolutional Neural Networks for Segmentation of Diffusely Abnormal
White Matter in Multiple Sclerosis. Proceedings of the American Association of
Physicists in Medicine Annual Meeting, 2022.
- Sajja B, Datta S, He R, et al. Unified approach
for multiple sclerosis lesion segmentation on brain MRI. Ann Biomed Eng.
2006;34(1):142-151.
- Ronneberger O, Fischer P, Brox T. U-Net:
Convolutional Networks for Biomedical Image Segmentation. arXiv. 2015;1505.04597.
- Papadaki E, Mastorodemos V, Panou T, et al. T2
Relaxometry Evidence of Microstructural Changes in Diffusely Abnormal White
Matter in Relapsing-Remitting Multiple Sclerosis and Clinically Isolated
Syndrome: Impact on Visuomotor Performance. J Magn Reson Imaging. 2021;54(4):1077-1087.