Conference PaperPDF Available

Cross-vendor and Cross-protocol harmonisation of diffusion MRI data: a comparative study

Authors:

Abstract and Figures

We present a comparison of five different methods that estimate mappings between scanners for diffusion MRI data harmonisation. The methods are evaluated on a dedicated dataset of the same subjects acquired on three distinct scanners with 'standard' and 'state-of-the-art' protocols, with the latter having higher spatial and angular resolution. Our results show that cross-vendor harmonisation and spatial/angular resolution enhancement of single-shell diffusion data sets can be performed reliably, although some challenges remain. The dataset is available upon request and can serve as a useful testbed for future method development in cross-site/cross-hardware and cross-vendor diffusion MRI harmonisation.
Content may be subject to copyright.
A preview of the PDF is not available
... We demonstrate our proposed method on the MICCAI Computational Diffusion MRI challenge dataset, [20][21][22] showing substantial improvement compared to a recently published baseline method. We also introduce technical improvements to the training of neural architectures on diffusion-weighted data, and discuss the limitations and error modes of our proposed method. ...
... 10,12,13 Further protocol differences arise between sites due to limitations of the available scanners, unavoidable changes or upgrades in scanners or protocols, or when combining data retrospectively from multiple studies; effects of variations in scanning protocols on derived measures include effects of voxel size, 11 b-values (the diffusion weightings used), 8,11 and angular resolution or q-space sampling 9,14-16 among other parameters. These problems were also examined by the MICCAI Computational Diffusion MRI 2017 and 2018 challenges, 20,21 which held an open comparison of methods for a supervised (paired) task. ...
... This requires explicitly paired data, in that the same brains must be scanned at all sites. These methods perform well empirically, as tested by the CDMRI challenge, 21 but require paired data in the training set. Our proposed method does not require paired data to train; however, in our opinion, best practice validation still requires paired data in the (holdout) test set. ...
Article
Full-text available
Purpose In the present work, we describe the correction of diffusion‐weighted MRI for site and scanner biases using a novel method based on invariant representation. Theory and Methods Pooled imaging data from multiple sources are subject to variation between the sources. Correcting for these biases has become very important as imaging studies increase in size and multi‐site cases become more common. We propose learning an intermediate representation invariant to site/protocol variables, a technique adapted from information theory‐based algorithmic fairness; by leveraging the data processing inequality, such a representation can then be used to create an image reconstruction that is uninformative of its original source, yet still faithful to underlying structures. To implement this, we use a deep learning method based on variational auto‐encoders (VAE) to construct scanner invariant encodings of the imaging data. Results To evaluate our method, we use training data from the 2018 MICCAI Computational Diffusion MRI (CDMRI) Challenge Harmonization dataset. Our proposed method shows improvements on independent test data relative to a recently published baseline method on each subtask, mapping data from three different scanning contexts to and from one separate target scanning context. Conclusions As imaging studies continue to grow, the use of pooled multi‐site imaging will similarly increase. Invariant representation presents a strong candidate for the harmonization of these data.
... We demonstrate our proposed method on CDMRI challenge dataset [31], showing substantial improvement compared to a recently published baseline. We further introduce technical improvements to the training of neural architectures on diffusion weighted data, and discuss the limitations and error modes of our proposed method. ...
... Further protocol differences arise between sites due to limitations of the available scanners, unavoidable changes or upgrades in scanners or protocols, or when combining data retrospectively from multiple studies; effects of scanning protocols on derived measures include effects of voxel size [26], b-values (the diffusion weightings used) [5,8,26], and angular resolution or q-space sampling [10,35,36,37] among other parameters. These problems were also looked at by the CDMRI 2017 and 2018 challenges [31], which held an open comparison of methods for a supervised (paired) task. ...
... This family of methods generally relies on high expressivecapacity function fitting (e.g., neural networks) to map directly between patches of pairs of images, assumably the "same" brain scanned at both sites. These methods perform well empirically, as tested by the CDMRI challenge [31], but require paired data in the training set. Our proposed method does not require paired data to train; however, in our opinion best practice validation still requires paired data in the (holdout) test-set. ...
Preprint
Pooled imaging data from multiple sources is subject to variation between the sources. Correcting for these biases has become incredibly important as the size of imaging studies increases and the multi-site case becomes more common. We propose learning an intermediate representation invariant to site/protocol variables, a technique adapted from information theory-based algorithmic fairness; by leveraging the data processing inequality, such a representation can then be used to create an image reconstruction that is uninformative of its original source, yet still faithful to the underlying structure. To implement this, we use a machine learning method based on variational auto-encoders (VAE) to construct scanner invariant encodings of the imaging data. To evaluate our method, we use training data from the 2018 CDMRI Challenge Harmonization dataset. Our proposed method shows improvements on independent test data relative to a recently published baseline method.
... In addition to ComBat, a number of harmonization approaches have recently been proposed at various stages of analysis (Tax et al., 2018;Zhu et al., 2018). Site differences can be accounted for at the time of overall group inference, such as with the random-effects regression level correction used here, or by using a meta-analysis approach in lieu of pooling data (Thompson et al., 2014). ...
Article
Full-text available
Brain imaging with diffusion-weighted MRI (dMRI) is sensitive to microstructural white matter (WM) changes associated with brain aging and neurodegeneration. In its third phase, the Alzheimer’s Disease Neuroimaging Initiative (ADNI3) is collecting data across multiple sites and scanners using different dMRI acquisition protocols, to better understand disease effects. It is vital to understand when data can be pooled across scanners, and how the choice of dMRI protocol affects the sensitivity of extracted measures to differences in clinical impairment. Here, we analyzed ADNI3 data from 317 participants (mean age: 75.4 ± 7.9 years; 143 men/174 women), who were each scanned at one of 47 sites with one of six dMRI protocols using scanners from three different manufacturers. We computed four standard diffusion tensor imaging (DTI) indices including fractional anisotropy (FADTI) and mean, radial, and axial diffusivity, and one FA index based on the tensor distribution function (FATDF), in 24 bilaterally averaged WM regions of interest. We found that protocol differences significantly affected dMRI indices, in particular FADTI. We ranked the diffusion indices for their strength of association with four clinical assessments. In addition to diagnosis, we evaluated cognitive impairment as indexed by three commonly used screening tools for detecting dementia and AD: the AD Assessment Scale (ADAS-cog), the Mini-Mental State Examination (MMSE), and the Clinical Dementia Rating scale sum-of-boxes (CDR-sob). Using a nested random-effects regression model to account for protocol and site, we found that across all dMRI indices and clinical measures, the hippocampal-cingulum and fornix (crus)/stria terminalis regions most consistently showed strong associations with clinical impairment. Overall, the greatest effect sizes were detected in the hippocampal-cingulum (CGH) and uncinate fasciculus (UNC) for associations between axial or mean diffusivity and CDR-sob. FATDF detected robust widespread associations with clinical measures, while FADTI was the weakest of the five indices for detecting associations. Ultimately, we were able to successfully pool dMRI data from multiple acquisition protocols from ADNI3 and detect consistent and robust associations with clinical impairment and age.
... Recently, methods based on deep learning have been developed for data harmonization [37,38,39]. While effective, deep learning approaches require training data, i.e., subjects scanned at multiple sites and with different scanners or protocols. ...
Article
Full-text available
Diffusion MRI is a powerful tool for non-invasive probing of brain tissue microstructure. Recent multi-center efforts in acquisition and analysis of diffusion MRI data significantly increase sample sizes and hence improve sensitivity and reliability in detecting subtle changes associated with development, aging, and diseases. However, discrepancies resulting from different scanner vendors, acquisition protocols, and image reconstruction algorithms can cause data incompatibility across imaging centers. In this paper, we introduce a modelfree method that is based on the method of moments (MoM) for direct harmonization of diffusion MRI data to reduce sitespecific variations. Our method directly harmonizes diffusionattenuated signal without the need to fit any diffusion model. Moreover, our method allows the explicit definition of wellbehaved mapping functions with properties such as invertibility, smoothness, and injectivity. We show that our method is effective in lowering variations of diffusion scalars of traveling human phantoms scanned at different sites from 1%–3% to less than 0.9% for fractional anisotropy (FA) and mean diffusivity (MD), and from 1%–2.5% to 0.3%–1.2% for generalized fractional anisotropy (GFA). We also demonstrate its ability in preserving individual differences and in increasing across-site consistency in tractography and white matter connectivity.
Chapter
Advances in diffusion MRI (dMRI) have led to discoveries of factors that affect brain microstructure and connectivity in health and disease. The small size of many neuroimaging studies led to concerns about poor reproducibility of research findings, and calls for the comparison and pooling of multi-cohort datasets to establish the consistency of reported effects. Across studies diffusion MRI protocols vary in spatial, angular and q-space resolution, b-value, as well as hardware used—all of which affect measured diffusion parameters. Efforts to compare and pool dMRI measures use meta- or mega- analytical techniques to compensate for these sources of variance. Meta-analytical methods gauge the consistency of effects, and mega-analytical methods involve mathematical or statistical transformations of the data. Here, we review some recent advances that allowed the diffusion community to create large scale population studies with greater rigor and generalizability than was previously attainable by individual studies.
Chapter
Diffusion imaging is an important method in the field of neuroscience, as it is sensitive to changes within the tissue microstructure of the human brain. However, a major challenge when using MRI to derive quantitative measures is that the use of different scanners, as used in multi-site group studies, introduces measurement variability. This can lead to an increased variance in quantitative metrics, even if the same brain is scanned. Contrary to the assumption that these characteristics are comparable and similar, small changes in these values are observed in many clinical studies, hence harmonization of the signals is essential. In this paper, we present a method that does not require additional preprocessing, such as segmentation or registration, and harmonizes the signal based on a deep learningresidual network. For this purpose, a training database is required, which consist of the same subjects, scanned on different scanners. The results show that harmonized signals are significantly more similar to the ground truth signal compared to no harmonization, but also improve in comparison to another deep learning method. The same effect is also demonstrated in commonly used metrics derived from the diffusion MRI signal.
Chapter
We present a summary of competition results in the multi-shell diffusion MRI harmonisation and enhancement challenge (MUSHAC). MUSHAC is an open competition intended to stimulate the development of computational methods that reduce scanner- and protocol-related variabilities in multi-shell diffusion MRI data across multi-site studies. Twelve different methods from seven research groups have been tested in this challenge. The results show that cross-vendor harmonization and enhancement can be performed by using suitable computational algorithms such as deep convolutional neural networks. Moreover, parametric models for multi-shell diffusion MRI signals also provide reliable performances.
  • Mirzaalian
Mirzaalian et al., NeuroImage 135:311-23,2016
  • Mirzaalian
Mirzaalian et al., Brain Imaging Behav. doi: 10.1007/s11682-016-9670-y,2017
  • Fortin
Fortin et al., NeuroImage 161:149-170,2017;
  • Sotiropoulos Andersson
Andersson and Sotiropoulos,NeuroImage 125:1063-1078,2016
  • Andersson
Andersson et al.,NeuroImage 20(2):870-888,2003
  • Glasser
Glasser et al.,Neuroimage 80:105-124,2013
  • Jenkinson
Jenkinson et al.,NeuroImage 17(2),825-841,2002
  • Klein
Klein et al.,IEEE TMI 29(1),196-205 2010
  • Irfanoglu
Irfanoglu et al.,Neuroimage 61(1),275-288,2012
  • Mirzaalian
Mirzaalian et al., Med Image Comput Comput Assist Interv. 9349:12-19,2015