Neuroimage Analysis Center

Neuroimage Analysis Center
"understanding the human brain through imaging"

fMRI Informatics Core

Introduction

The Functional MRI Informatics Core is a technology research and development (TRD) project that is focused on the development and application of new statistical and information theoretic image analysis methods that are aimed at population studies.

Under the leadership of Sandy Wells, the scientific director of the NAC, the project focuses on the development of statistical methodology in the following areas:

  1. Pooling fMRI activations across subjects and scanning sites,
  2. Multi-modality analysis of brain activation, and
  3. Registering populations of subjects together.

Because it is non-invasive, safe, and has relatively high spatial resolution, fMRI has become an important modality for the study of brain function in neuroscience, disease research, and neurosurgery. In order to facilitate such research, we focus on the analysis of differences in brain activation within and across groups of subjects and/or patients.

Despite its advantages, fMRI has low signal to noise ratio and poor temporal resolution in relation to neuronal time scales. Because of this, the analysis of fMRI data is interesting from the point of view of signal processing and applied statistics.

The aims of the TRD project are describe in more detail below.

Aim 1: Pooling fMRI activations across subjects and sites

In order to gain statistical power in determining group differences, it is important to combine, or "pool", data across multiple subjects. This is a complex problem in practice, partly because of the idiosyncrasies of the fMRI signal, and partly because of the the neuroanatomical variability in the population.

We have previously developed and used STAPLE for a preliminary analysis in the context of fBIRN, a large multi site fMRI scanning project. This analysis depended on thresholding the fMRI signal for activation strength, which potentially loses some information. We are currently developing an alternative Bayesian methodology to address the pooling problem while avoiding thresholding.

The analysis of activation in populations is also being pursued in a local collaboration with Prof. Polina Golland at MIT, in a line of research that focuses on clustering the time-courses of fMRI signals in voxels.

Aim 2: Multi-modality analysis of brain activation

Our neurosurgical colleagues, including Dr. Alex Golby, sometimes use other functional modalities for observing brain activity, including electrical (EEG) and magnetic (MEG) modalities, which have higher temporal resolution than fMRI, but lower spatial resolution. Our research in this area is aimed at inference on activation that is based on joint, but not necessarily simultaneous, observations of multiple modalities. Ideally, we hope to approach the spatial resolution of fMRI and temporal resolution of the others.

In other multimodal analysis, we study fMRI in conjunction with intraoperative electrical stimulation (DECS), which is a standard method of intraoperative brain mapping. One of the major goals of this internal collaboration with our neurosurgical colleagues is to evaluate the validity of fMRI as a surgical mapping tool.

The figure illustrates the combination of fMRI and DECS in a surgical case.
The figure illustrates the combination of fMRI and DECS in a surgical case.

As mentioned above, the neuroanatomical variability in the population is a thorny issue that complicates the study of brain activation, as one frequently would like to examine activity in corresponding structures. This problem can be partially addressed by normalizing subject scans to an anatomical atlas, but this technology is currently somewhat limited, and the residual inaccuracy of the process is frequently addressed by blurring the fMRI signals in space, which causes some loss of effective spatial resolution. Additionally, the variability of the location of functional areas with respect to the individual brain morphology is not currently well understood.

To gain traction with these difficulties we plan to pursue an analysis that will, in the statistical sense, jointly model variability on the spatial aspects of structure and function. An initial goal of this work is to quantify the amount of spatial uncertainty in the prediction of the location of a functional area in an individual, based on statistical knowledge of the population.

Aim 3: Registering populations of subjects together

As outlined above, neuroanatomical variability in the population is a substantial complexity in the analysis of brain activation, and one of the main techniques for accommodating the variability is the use of anatomical atlases and registration techniques. In this area, we have been pursuing the development of an entropy based group registration method called "congealing". This is joint work with Erik Learned-Miller and Lilla Zollei. The approach forms an atlas, or standard space, by a group registration process that is inherently unbiased.

Group Registration of twenty infant MRI scans - Left: before registration, Right: After Registration. 200 Iterations of Stochastic Gradient Descent was use with sampling from 1000 voxel locations
Group Registration of twenty infant MRI scans - Left: before registration, Right: After Registration. 200 Iterations of Stochastic Gradient Descent was use with sampling from 1000 voxel locations

Background

The current research activity of the project builds on a our previous research in the areas of segmentation and registration.

Segmentation

Our group developed the original "EM Segmenter", which was the first algorithm that could successfully segment MRI into white matter and gray matter in an automatic fashion. The method has evolved over the years, incorporating contributions by Tina Kapur in area of Markov Random Field to fight noise, and by Kilian Pohl with the use of anatomical atlas registration.

The figure illustrates the evolution of a simultaneous segmentation and atlas registration.
The figure illustrates the evolution of a simultaneous segmentation and atlas registration.

Registration

We also participated in the original development of the maximization of mutual information (MI) method of image registration, which has become the default method of choice for problems of multi-modality image registration.

Subsequent developments in this area have included collaborative work with Albert Chung on the KLD approach, which provides a of incorporating domain-specific modeling into registration , and with Lilla Zollei on the Dirichlet approach, which provides a principled way of controlling the domain modeling.


Bibliography


  1. Balci S, Golland P, Shenton M, Wells W. Free-Form B-spline Deformation Model for Groupwise Registration. Workshop: Statistical Registration: Pair-wise and Group-wise Alignment and Atlas Formation. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2007;10(WS):23-30.
  2. HM Chan, A Chung, S Yu, A Norbash, W Wells. Multi-modal image registration by minimizing Kullback-Leibler distance between expected and observed joint class histograms. IEEE Conference on Computer Vision and Pattern Recognition, 2003.
  3. E. Cosman, J. Fisher, W. Wells. Exact MAP Activity Detection in fMRI using a GLM with an Ising Spatial Prior Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2004;7(Pt 2):703-710.
  4. E R Cosman, Jr. and WM Wells. Bayesian Population Modeling of Effective Connectivity. Inf Process Med Imaging. 2005;19:39-51.
  5. A Fan, J Fisher, M Cedolin, W Wells. A Unified Variational Approach to Denoising and Bias Correction in MR. Inf Process Med Imaging. 2003 Jul;18:148-59.
  6. Fan A, Fisher J, Wells W, Levitt J, Willsky A. MCMC Curve Sampling for Image Segmentation. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2007;10(Pt 2):477-485.
  7. R. Gan, J. Wu, A. Chung, S. Yu, W. Wells. Multiresolution image registration based on Kullback-Leibler distance. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2004;7(Pt 1):599-606.
  8. S Larsen, W Wells, A Golby, I Talos, R Kikinis. Quantitative Comparison of Function MRI and Direct Electro-Cortical Stimulation for Functional Mapping. Abstract, Human Brain Mapping Conference, Budapest, 2004.
  9. Larsen S, Kikinis R, Talos IF, Weinstein D, Wells W, Golby A. Quantitative comparison of functional MRI and direct electrocortical stimulation for functional mapping. Int J Med Robot. 2007 Sep;3(3):262-70.
  10. Maddah M, Wells W, Warfield S, Westin CF, Grimson E. Probabilistic Clustering and Quantitative Analysis of White Matter Fiber Tracts. Inf Process Med Imaging. 2007;20:372-83.
  11. Mewes A, Zollei L, Huppi P, Als H, McAnulty G, Inder T, Wells W, Warfield S. Displacement of Brain Regions in Preterm Infants and Non-Synostotic Dolichocephaly Investigated by MRI. Neuroimage 36(4), 1074-1085, 2007.
  12. K. Pohl, S. Warfield, R. Kikinis, W. Grimson, W. Wells. Coupling Statistical Segmentation and PCA Shape Modeling. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2004;7(Pt 2):151-159.
  13. K Pohl, J Fisher, R Kikinis, W.E.L. Grimson, and W Wells. Shape Based Segmentation of Anatomical Structures in Magnetic Resonance Images. Proc. ICCV 2005: Computer Vision for Biomedical Image Applications: Current Techniques and Future Trends, An International Conference on Computer Vision Workshop, Beijing, China, Springer-Verlag, vol. 3765 of Lecture Notes in Computer Science, 2005.
  14. K Pohl, J Fisher, J Levitt, M Shenton, R Kikinis, W.E.L. Grimson, and W Wells. A Unifying Approach to Registration, Segmentation, and Intensity Correction. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2005;8(Pt 1):310-8.
  15. Pohl K M, Fisher J, Grimson WEL, Kikinis R, Wells W. A Bayesian Model for Joint Segmentation and Registration. Neuroimage 31(1), 228-239, 2006.
  16. Pohl K, Fisher J, Shenton M, McCarley R, Grimson WEL, Kikinis R, Wells W. Logarithm odds maps for shape representation. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2006;9(Pt 1):955-963.
  17. Pohl K, Bouix S, Nakamura M, Rohlfing T, McCarley R, Kikinis R, Grimson WEL, Shenton M, Wells W. A Hierarchical Algorithm for MR Brain Image Parcellation. EEE Transactions on Medical Imaging. 2007 Sept;26(9):1201-1212..
  18. Pohl K, Fisher J, Grimson E, Kikinis R, Wells W. Using the Logarithm of Odds to Define a Vector Space on Probabilistic Atlases. Med Image Anal. 2007 Oct;11(5):465-77.
  19. Pohl K, Kikinis R, Wells W. Active Mean Fields: Solving the Mean Field Approximation in the Level Set Framework. Inf Process Med Imaging. 2007;20:26-37.
  20. S Soman, A Chung, W Grimson, W Wells. Echoplanar Image (EPI) to MRI Rigid Rigid Registration by Minimizing Kullback-Leibler Distance. Second International Workshop on Biomedical Image Registration 2003.
  21. IF Talos, L Odonnell, CF Westin, S Warfield, W Wells, SS Yoo, L Panych, A Golby, H Mamata, S Maier, P Ratiu, C Guttmann, P McL Black, F Jolesz and R Kikinis. Diffusion Tensor and Functional MRI Fusion with Anatomical MRI for Image Guided Neurosurgery. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2006;6(Pt 1):407-415.
  22. Tsai A, Wells WM, Tempany C, Grimson E, Willsky A. Mutual Information in a Coupled Multi-Shape Model for Medical Image Segmentation. Medical Image Analysis 2004: 429-445.
  23. A Tsai, W Wells, S Warfield, A Willsky. An EM Algorithm for Shape Classification Based on Level Sets. Med Image Anal. 2005 Oct;9(5):491-502.
  24. A. Tsai, W. Wells, S. Warfield, A. Willsky. Level Set Methods in an EM Framework for Shape Classification and Estimation. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2004;7(Pt 1):1-9.
  25. S Warfield, F Talos, C Kemper, E Cosman, A Tei, M Ferrant, B Macq, W Wells, P McL Black, F Jolesz, R Kikinis. Augmenting Intraoperative MRI with Preoperative fMRI and DTI by Biomechanical Siulation of Brain Deformation. SPIE Conference on Medical Imaging, San Diego, Ca, 2003.
  26. Warfield S, Zou KH, Wells WM. Simultaneous Truth and Performance Level Estimation (STAPLE): An Algorithm for the Validation of Image Segmentation. IEEE Trans Med Imag 2004; 23:903-921.
  27. Warfield S, Zou K, Wells W. Validation of Image Segmentation by Estimating Rater Bias and Variance. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2006;9(Pt 2):839-47.
  28. L Zollei, J. Fisher, W. Wells. A Unified Statistical and Information Theoretic Framework for Multi-modal Image Registration. Inf Process Med Imaging. 2003 Jul;18:366-77.
  29. L Zollei, L Paych, E Grimson, W Wells. Exploratory Identification of Cardiac Noise in fMRI Images. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2006;6(Pt 1):475-483.
  30. L Zollei, E Learned-Miller, E Grimson, W Wells. Efficient Population Registration of 3D Data. Proc. ICCV 2005: Computer Vision for Biomedical Image Applications: Current Techniques and Future Trend, An International Conference on Computer Vision Workshop, Beijing, China, Springer-Verlag, LNCS 3765, 2005. (Received Best Paper Award).
  31. L Zollei, J Fisher, W Wells. An Introduction to Statistical Methods of Medical Image Registration. Mathematical Models in Computer Vision: The Handbook, Springer (2005).
  32. L Zollei, W Wells. Multi-modal Image Registration Using Dirichlet-encoded Prior Information. Proceedings of the 3rd International Workshop on Biomedical Image Registration, WBIR 2006, LNCS 4057, pp. 34–42, 2006.
  33. Zollei L, Jenkinson M, Timoner S, Wells W. A Marginalized MAP Approach and EM Optimization for Pair-Wise Registration. Inf Process Med Imaging. 2007;20:662-74.
  34. Zollei L, Shenton M, Wells W, Pohl K. The Impact of Atlas Formation Methods on Atlas-Guided Brain Segmentation. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2007;10(WS):39-46.
  35. K. Zou, D. Greve, M. Wang, S. Pieper, S. Warfield, N. White, M. Vangel, R. Kikinis, W. Wells, F. BIRN. A prospective multi-Institutional study of the reproducibility of fMRI: a preliminary report from the biomedical informatics research network. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2004;7(Pt 2):769-776.
  36. K Zou, D Greve, M Wang, S Pieper, S Warfield, N White, S Manandhar, G Brown, M Vangel, R Kikinis, and W Wells. Reproducibility of functional MR imaging: Preliminary results of a prospective multi-institutional study by the Biomedical Informatics Research Group. Radiology. 2005 Dec;237(3):781-9.
  37. Zou KH, Wells W, Kikinis R, Warfield SK. Three validation metrics for automated probabilistic image segmentation of brain tumors. Stat Med. 2004 Apr 30;23(8):1259-82.
Retrieved from "http://nac.spl.harvard.edu/pages/Research_Cores/fMRIInformatics"