Image Analysis
Anastasia Yendiki1

1Athinoula A. Martinos Center for Biomedical Imaging, Harvard Medical School and Massachusetts General Hospital

Synopsis

Combining MRI data collected at multiple sites allows researchers to achieve the large sample sizes required to detect subtle disease effects, although at the expense of increased inhomogeneity in the data due to differences in acquisition hardware and software. This presentation will address what is known about the reproducibility of biomarkers derived from structural, functional, and diffusion MRI data across scanners from different vendors, as well as image analysis strategies that have been proposed to mitigate the effects of scanner-related differences.

Target audience

This presentation is targeted at clinicians and scientists who want to perform studies with data acquired on multiple MRI platforms, as well as image analysis researchers interested in developing methods to facilitate such studies.

Outcome/objectives

The goal of this presentation is to familiarize the audience with the effects of MRI scanner differences on the biomarkers that are typically extracted from structural, functional, and diffusion MRI data, as well as some of the image analysis methods that have been proposed to mitigate the effects of scanner differences retrospectively.

Purpose

Acquiring MRI data concurrently at multiple sites allows researchers to achieve the large sample sizes required to detect subtle disease effects. However, differences in acquisition hardware and sequences across MRI scanner vendors introduce variability that can reduce the gain in statistical power afforded by the increased sample size. Compensating for scanner-dependent image distortions, such as B0 or B1 field inhomogeneities, gradient field nonlinearities, and eddy currents, can decrese the variability across images collected on different scanners, but not eliminate it completely. Calibration scans with phantoms can be used to quantify the differences in signal-to-noise ratio and distortions across platforms, but it is not easy to determine how a difference observed in phantom data would translate to the types of structural and functional biomarkers that are extracted from human data. Therefore there is a need for studies that assess the effects of scanner differences on these biomarkers, and for methods to analyze data in the presence of such differences.

Results on across-vendor reproducibility of MRI biomarkers

Several studies have quantified the reproducibility of structural, functional, and diffusion MRI biomarkers across scanners from different vendors. Early studies in structural MRI showed that compensating for gradient nonlinearities reduces differences in image intensities [1], and compensating for B1 inhomogeneities reduces variability in tensor-based morphometry [2]. The choice of sequence was found to be important for the across-scanner reproducibility of surface-based morphometric measures, including cortical thickness [3] and the volumes of subcortical structures [4]. Significant effects of scanner differences across vendors have been reported in voxel-based morphometry [5-6]. For task-based functional MRI data collected at sites that had scanners from different vendors, the inter-scanner variability was found to be an order of magnitude lower than the inter-individual variability [7-9]. For resting-state functional MRI, across-scanner differences were sometimes significant, but varied depending on whether functional correlation coefficients were computed using an a priori seed region or a data-driven approach [10]. In diffusion MRI, reproducibility studies that involved multiple vendors showed small but statistically significant inter-scanner effects in anisotropy and diffusivity measures [11-13].

Methods for image analysis in the presence of scanner differences

Discrepancies between images collected on different MRI systems can pose a problem for atlas-based image analyses, where an algorithm may be trained on data acquired on one system and applied to test data acquired on other systems, or in multi-site studies, where the test data itself is acquired on different systems. In atlas-based analyses, it is important to train algorithms on image features that are not sensitive to the acquisition. An example is the position of brain structures relative to each other, which can be used in atlas-based structural MRI segmentation [14,15]. The relative positions of the structures are invariant to changes in the acquisition, even if, say, the histograms of the image intensities of voxels in each structure are not. We have used the positions of white-matter pathways relative to their surrounding anatomical structures as features for the atlas-based reconstruction of these pathways with diffusion MRI tractography [16], and we have recently shown that this approach is robust to acquisition differences between the training and test data [17].

In multi-site studies that have to combine MRI data collected on different systems, several strategies could be applied to reduce the effects of acquisition-related differences on downstream analyses. Including a forward model of the image intensities given a set of intrinsic tissue properties can make structural MRI segmentation invariant to the acquisition [18]. An approach known as "image synthesis" has been proposed to generate an estimate of a missing image for a test subject that has an image from a different modality, given pairs of images from both modalities for a set training subjects. Image synthesis methods have been applied, for example, to generate data with a structural MRI contrast that is appropriate for image segmentation for a subject who has data with a different contrast that is suboptimal for this purpose [19-21], or to upsample low-resolution diffusion MRI data given a set of high-resolution training data acquired on a different scanner [22]. Such methods may be useful for harmonizing images from multi-site studies. An alternative to harmonizing the images themselves is to do so for the spherical harmonic expansion coefficients of diffusion MRI data acquired on different sites [23]. Finally, approaches that do not rely on image harmonization include performing independent component analysis to identify scanner-related components [24] or otherwise modeling site variability in statistical analyses [25,26].

Discussion

Several findings of prior reproducibility studies reveal the effect of scanner-related differences on the biomarkers computed from structural, functional, and diffusion MRI data, but also the impact that the choice of image analysis methods can have on across-vendor reproducibility. Several analysis methods proposed in the literature hold promise for mitigating the effects of acquisition-related variability in the images, but further research is needed to assess their performance in the presence of significant across-vendor differences.

Conclusion

The proliferation of publicly available MRI data sets collected across multiple platforms is making the problem of combining these data sets increasingly relevant. As acquisition technologies evolve, with higher field strengths, higher gradient strengths, and accelerated sequences becoming more widely available, further studies will be necessary to assess the across-platform reproducibility of MRI biomarkers obtained with these technologies.

Acknowledgements

The author is grateful to Dr. Bruce Fischl for helpful discussions on the topic of this presentation.

References

  1. J. Jovicich, S. Czanner, D. Greve, E. Haley, A. van der Kouwe, R. Gollub, D. Kennedy, F. Schmitt, G. Brown, J. Macfall, B. Fischl, and A. Dale, “Reliability in multi-site structural MRI studies: effects of gradient non-linearity correction on phantom and human data.,” Neuroimage, vol. 30, no. 2, pp. 436–443, Apr. 2006.
  2. A. D. Leow, A. D. Klunder, C. R. Jack, A. W. Toga, A. M. Dale, M. A. Bernstein, P. J. Britson, J. L. Gunter, C. P. Ward, J. L. Whitwell, B. J. Borowski, A. S. Fleisher, N. C. Fox, D. Harvey, J. Kornak, N. Schuff, C. Studholme, G. E. Alexander, M. W. Weiner, P. M. Thompson, ADNI Preparatory Phase Study, “Longitudinal stability of MRI for mapping brain change using tensor-based morphometry.,” Neuroimage, vol. 31, no. 2, pp. 627–640, Jun. 2006.
  3. X. Han, J. Jovicich, D. Salat, A. van der Kouwe, B. Quinn, S. Czanner, E. Busa, J. Pacheco, M. Albert, R. Killiany, P. Maguire, D. Rosas, N. Makris, A. Dale, B. Dickerson, and B. Fischl, “Reliability of MRI-derived measurements of human cerebral cortical thickness: the effects of field strength, scanner upgrade and manufacturer.,” Neuroimage, vol. 32, no. 1, pp. 180–194, Aug. 2006.
  4. J. Jovicich, S. Czanner, X. Han, D. Salat, A. van der Kouwe, B. Quinn, J. Pacheco, M. Albert, R. Killiany, D. Blacker, P. Maguire, D. Rosas, N. Makris, R. Gollub, A. Dale, B. C. Dickerson, and B. Fischl, “MRI-derived measurements of human subcortical, ventricular and intracranial brain volumes: Reliability effects of scan sessions, acquisition sequences, data analyses, scanner upgrade, scanner vendors and field strengths.,” Neuroimage, vol. 46, no. 1, pp. 177–192, May 2009.
  5. H. Pardoe, G. S. Pell, D. F. Abbott, A. T. Berg, and G. D. Jackson, “Multi-site voxel-based morphometry: methods and a feasibility demonstration with childhood absence epilepsy.,” Neuroimage, vol. 42, no. 2, pp. 611–616, Aug. 2008.
  6. H.-J. Huppertz, J. Kröll-Seger, S. Kloppel, R. E. Ganz, and J. Kassubek, “Intra- and interscanner variability of automated voxel-based volumetry based on a 3D probabilistic atlas of human cerebral structures.,” Neuroimage, vol. 49, no. 3, pp. 2216–2224, Feb. 2010.
  7. A. Yendiki, D. N. Greve, S. Wallace, M. Vangel, J. Bockholt, B. A. Mueller, V. Magnotta, N. Andreasen, D. S. Manoach, and R. L. Gollub, “Multi-site characterization of an fMRI working memory paradigm: reliability of activation indices.,” Neuroimage, vol. 53, no. 1, pp. 119–131, Oct. 2010.
  8. G. G. Brown, D. H. Mathalon, H. Stern, J. Ford, B. Mueller, D. N. Greve, G. McCarthy, J. Voyvodic, G. Glover, M. Diaz, E. Yetter, I. B. Ozyurt, K. W. Jorgensen, C. G. Wible, J. A. Turner, W. K. Thompson, S. G. Potkin, Function Biomedical Informatics Research Network, “Multisite reliability of cognitive BOLD data.,” Neuroimage, vol. 54, no. 3, pp. 2163–2175, Feb. 2011.
  9. J. K. Forsyth, S. C. McEwen, D. G. Gee, C. E. Bearden, J. Addington, B. Goodyear, K. S. Cadenhead, H. Mirzakhanian, B. A. Cornblatt, D. M. Olvet, D. H. Mathalon, T. H. McGlashan, D. O. Perkins, A. Belger, L. J. Seidman, H. W. Thermenos, M. T. Tsuang, T. G. M. van Erp, E. F. Walker, S. Hamann, S. W. Woods, M. Qiu, and T. D. Cannon, “Reliability of functional magnetic resonance imaging activation during working memory in a multi-site study: analysis from the North American Prodrome Longitudinal Study.,” Neuroimage, vol. 97, pp. 41–52, Aug. 2014.
  10. J. Jovicich, L. Minati, M. Marizzoni, R. Marchitelli, R. Sala-Llonch, D. Bartrés-Faz, J. Arnold, J. Benninghoff, U. Fiedler, L. Roccatagliata, A. Picco, F. Nobili, O. Blin, S. Bombois, R. Lopes, R. Bordet, J. Sein, J.-P. Ranjeva, M. Didic, H. Gros-Dagnac, P. Payoux, G. Zoccatelli, F. Alessandrini, A. Beltramello, N. Bargalló, A. Ferretti, M. Caulo, M. Aiello, C. Cavaliere, A. Soricelli, L. Parnetti, R. Tarducci, P. Floridi, M. Tsolaki, M. Constantinidis, A. Drevelegas, P. M. Rossini, C. Marra, P. Schönknecht, T. Hensch, K.-T. Hoffmann, J. P. Kuijer, P. J. Visser, F. Barkhof, G. B. Frisoni, PharmaCog Consortium, “Longitudinal reproducibility of default-mode network connectivity in healthy elderly participants: A multicentric resting-state fMRI study.,” Neuroimage, vol. 124, no. Pt A, pp. 442–454, Jan. 2016.
  11. M. Cercignani, R. Bammer, M. P. Sormani, F. Fazekas, and M. Filippi, “Inter-sequence and inter-imaging unit variability of diffusion tensor MR imaging histogram-derived metrics of the brain in healthy volunteers.,” AJNR Am J Neuroradiol, vol. 24, no. 4, pp. 638–643, Apr. 2003.
  12. E. Pagani, J. G. Hirsch, P. J. W. Pouwels, M. A. Horsfield, E. Perego, A. Gass, S. D. Roosendaal, F. Barkhof, F. Agosta, M. Rovaris, D. Caputo, A. Giorgio, J. Palace, S. Marino, N. De Stefano, S. Ropele, F. Fazekas, and M. Filippi, “Intercenter differences in diffusion tensor MRI acquisition.,” J Magn Reson Imaging, vol. 31, no. 6, pp. 1458–1468, Jun. 2010.
  13. V. A. Magnotta, J. T. Matsui, D. Liu, H. J. Johnson, J. D. Long, B. D. Bolster, B. A. Mueller, K. Lim, S. Mori, K. G. Helmer, J. A. Turner, S. Reading, M. J. Lowe, E. Aylward, L. A. Flashman, G. Bonett, and J. S. Paulsen, “Multicenter reliability of diffusion tensor imaging.,” Brain Connect, vol. 2, no. 6, pp. 345–355, 2012.
  14. B. Fischl, D. H. Salat, E. Busa, M. Albert, M. Dieterich, C. Haselgrove, A. van der Kouwe, R. Killiany, D. Kennedy, S. Klaveness, A. Montillo, N. Makris, B. Rosen, and A. M. Dale, “Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain.,” Neuron, vol. 33, no. 3, pp. 341–355, Jan. 2002.
  15. B. Fischl, A. van der Kouwe, C. Destrieux, E. Halgren, F. Ségonne, D. H. Salat, E. Busa, L. J. Seidman, J. Goldstein, D. Kennedy, V. Caviness, N. Makris, B. Rosen, and A. M. Dale, “Automatically parcellating the human cerebral cortex.,” Cereb Cortex, vol. 14, no. 1, pp. 11–22, Jan. 2004.
  16. A. Yendiki, P. Panneck, P. Srinivasan, A. Stevens, L. Zöllei, J. Augustinack, R. Wang, D. Salat, S. Ehrlich, T. Behrens, S. Jbabdi, R. Gollub, and B. Fischl, “Automated probabilistic reconstruction of white-matter pathways in health and disease using an atlas of the underlying anatomy.,” Front Neuroinform, vol. 5, p. 23, 2011.
  17. A. Yendiki, I. Filipiak, E. Saliba, L. Barantin, C. Destrieux, H. Dupuis, M. Cottier, J. Owen, Y. Ou, A. Varjabedian, C. Jaimes, L. Zollei, TRACULINA: Automated probabilistic tractography with anatomical priors in the infant brain, 22nd Annual Meeting of the Organization for Human Brain Mapping, 2016.
  18. B. Fischl, D. H. Salat, A. J. W. van der Kouwe, N. Makris, F. Ségonne, B. T. Quinn, and A. M. Dale, “Sequence-independent segmentation of magnetic resonance images,” Neuroimage, vol. 23, pp. S69–S84, Jan. 2004.
  19. D. H. Ye, D. Zikic, B. Glocker, A. Criminisi, and E. Konukoglu, “Modality Propagation: Coherent Synthesis of Subject-Specific Scans with Data-Driven Regularization,” in MICCAI, vol. 8149, no. 76, pp. 606–613, 2013.
  20. S. Roy, A. Carass, and J. L. Prince, “Magnetic Resonance Image Example-Based Contrast Synthesis.,” IEEE Trans Med Imaging, vol. 32, no. 12, pp. 2348–2363, Dec. 2013.
  21. A. Jog, A. Carass, S. Roy, D. L. Pham, and J. L. Prince, “MR image synthesis by contrast learning on neighborhood ensembles.,” Med Image Anal, vol. 24, no. 1, pp. 63–76, Aug. 2015.
  22. D. C. Alexander, D. Zikic, A. Ghosh, R. Tanno, V. Wottschel, J. Zhang, E. Kaden, T. B. Dyrby, S. N. Sotiropoulos, H. Zhang, and A. Criminisi, “Image quality transfer and applications in diffusion MRI.,” Neuroimage, vol. 152, pp. 283–298, Mar. 2017.
  23. H. Mirzaalian, L. Ning, P. Savadjiev, O. Pasternak, S. Bouix, O. Michailovich, S. Karmacharya, G. Grant, C. E. Marx, R. A. Morey, L. A. Flashman, M. S. George, T. W. McAllister, N. Andaluz, L. Shutter, R. Coimbra, R. D. Zafonte, M. J. Coleman, M. Kubicki, C.-F. Westin, M. B. Stein, M. E. Shenton, and Y. Rathi, “Multi-site harmonization of diffusion MRI data in a registration framework.,” Brain Imaging Behav, Feb. 2017.
  24. J. Chen, J. Liu, V. D. Calhoun, A. Arias-Vasquez, M. P. Zwiers, C. N. Gupta, B. Franke, and J. A. Turner, “Exploration of scanning effects in multi-site structural MRI studies.,” J. Neurosci. Methods, vol. 230, pp. 37–50, Jun. 2014.
  25. G. Salimi-Khorshidi, S. M. Smith, J. R. Keltner, T. D. Wager, T. E. Nichols, “Meta-analysis of neuroimaging data: A comparison of image-based and coordinate-based pooling of studies.,” vol. 45, no. 3, pp. 810–823, Apr. 2009.
  26. B. Zhou, A. Konstorum, T. Duong, K. H. Tieu, W. M. Wells, G. G. Brown, H. S. Stern, and B. Shahbaba, “A hierarchical modeling approach to data analysis and study design in a multi-site experimental fMRI study.,” Psychometrika, vol. 78, no. 2, pp. 260–278, Apr. 2013.
Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)