Evaluating the impact of denoising on diffusion MRI-based tractometry on glaucoma patients
Poster No:
1591
Submission Type:
Abstract Submission
Authors:
Daiki Taguma1,2, Shumpei Ogawa3, Hiromasa Takemura1,2,4
Institutions:
1National Institute for Physiological Sciences, Okazaki, Japan, 2The Graduate Institute of Advanced Studies, SOKENDAI, Hayama, Japan, 3The Jikei University School of Medicine, Tokyo, Japan, 4Center for Information and Neural Networks (CiNet), Advanced ICT Research Institute, NICT, Suita, Japan
First Author:
Daiki Taguma
National Institute for Physiological Sciences|The Graduate Institute of Advanced Studies, SOKENDAI
Okazaki, Japan|Hayama, Japan
National Institute for Physiological Sciences|The Graduate Institute of Advanced Studies, SOKENDAI
Okazaki, Japan|Hayama, Japan
Co-Author(s):
Hiromasa Takemura
National Institute for Physiological Sciences|The Graduate Institute of Advanced Studies, SOKENDAI|Center for Information and Neural Networks (CiNet), Advanced ICT Research Institute, NICT
Okazaki, Japan|Hayama, Japan|Suita, Japan
National Institute for Physiological Sciences|The Graduate Institute of Advanced Studies, SOKENDAI|Center for Information and Neural Networks (CiNet), Advanced ICT Research Institute, NICT
Okazaki, Japan|Hayama, Japan|Suita, Japan
Introduction:
Diffusion MRI (dMRI)-based tractometry is the sole method available for measuring tissue properties of white matter tracts in living humans (Jones et al., 2005; Yendiki et al., 2011; Yeatman et al., 2012). While this method offers high test-retest reliability (Kruper et al., 2021), dMRI data often suffers from measurement noise such as thermal noise limiting image quality. Several methods have been proposed to reduce noise in dMRI data (Veraart et al., 2016; Fadnavis et al., 2020). However, while these denoising methods improve overall image quality, it is unclear how much it affects the sensitivity of the tractometry-based measurement to identify tissue changes caused by disorders, without evaluating the empirical dataset. To address this, we assessed the impact of denoising on dMRI measurements along the optic tract of patients with glaucoma, a disorder that damages the retinal ganglion cells (RGCs) and likely affects tissue properties of the optic tract, which is composed of axons from RGCs (Quigley et al., 1989). Additionally, we also examined the effects of denoising on tractometry results for the optic radiation, which is not a part of RGCs but is often reported to be affected by glaucoma (Ogawa et al., 2022).
Methods:
We re-analyzed the dMRI dataset presented in previous work (Ogawa et al., 2022), including 17 glaucoma patients and 30 healthy controls. We compared dMRI data with and without denoising, which is implemented as the "dwidenoise" command in MRTrix3 (Veraart et al., 2016). dMRI data is then corrected for susceptibility-induced and eddy-current distortion using FSL. We identified the optic tract and optic radiation from preprocessed dMRI data using tractography methods described in previous work (Ogawa et al., 2022) and then evaluated tissue properties along these tracts using the AFQ MATLAB toolbox (Yeatman et al., 2012). Specifically, we evaluated the impact of denoising for intra-cellular volume fraction (ICVF), which is a dMRI-based measurement of microstructural properties (Zhang et al., 2012). Finally, we evaluated scan-rescan reproducibility of ICVF along the optic tract and optic radiation, to assess the impact on denoising for tractometry reproducibility.
Results:
Based on a visual inspection of the images, denoising improved the image quality of dMRI data, making tissue borders between gray matter and white matter more distinct. In addition, we found that glaucoma patients showed significantly lower ICVF compared with controls along the optic tract and optic radiation in both the data with and without denoising. Importantly, we found that denoising has no or little impact on the tractometry results. We found that the degree of abnormality in ICVF of glaucoma patients did not significantly differ between data with and without denoising, in both the optic tract and optic radiation (Figure 1; BF10 ≈ 0.3, 0.6 in each case). Finally, we also found that scan-rescan reproducibility of ICVF is comparable between data with and without denoising in both the optic tract and optic radiation (Figure 2).
Conclusions:
While denoising improved image quality, we found that it had minimal or no impact on sensitivity to identify tissue changes caused by glaucoma and on measurement reproducibility. These results are consistent with a recent study evaluating the impact of denoising in dMRI data of the spinal cord (Schilling et al., 2023). One possibility is that denoising may improve the fitting accuracy of voxelwise models to the dMRI signal by reducing measurement noise, however, resulting tissue properties measurement, such as ICVF, remain unchanged. Consequently, tractometry results do not seem to be affected. Therefore, we conclude that at present, we did not find evidence that denoising makes an actual impact on tractometry results on tissue changes caused by the retinal disorder.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis 1
Keywords:
Tractography
Vision
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Provide references using author date format
Jones et al. (2005) Magn Reson Med., 53(6), 1462-1467.
Kruper et al. (2021) Aperture Neuro,1(1).
Ogawa et al. (2022) Invest. Ophthalmol. Vis. Sci., 63(2), 29.
Quigley et al. (1989) Am. J. Ophthalmol., 107(5), 453-464.
Schilling et al. (2023) NeuroImage, 266, 119826.
Veraart et al. (2016) NeuroImage, 142, 394-406.
Yeatman et al. (2012) PLOS ONE, 7(11), e49790.
Yendiki et al. (2011) Front. Neuroinf., 5, 10815.
Zhang et al. (2012) NeuroImage, 61(4), 1000-1016.