New approaches to study connectivity, causality and patterns in functional MRI

Image courtesy of Martin Ystad, Tome Eichele, Erlend Hodneland, and Judit Haasz

Currently I am in NTU - Singapore Center for Computation Intelligence lab - school of Computer Science, doing my PhD under Dr. Vitali Zagorodnov. My work revolves around finding new machine learning approaches to study networks within the human brain and understand causality and patterns. Here is my research proposal...

Introduction and Motivation

Clinical applications always lag behind the theoretical developments. Sometimes this is due to improvements to well established algorithms being too marginal to incite a shift from existing clinical methods. In other cases, the new algorithms are simply too general, i.e. developed without reference to any specific application, and thus need substantial effort to adapt them to clinical application. Our research is motivated by several clinical applications in the area of cognitive neuroscience. The current trend in these (and many other) clinical applications is a slowly shifting focus from univariate to multivariate analysis to establish connectivity/ causality between medical measurements and clinical/behavioural observations.  We aim to contribute to this trend by bridging the gap between these applications and relatively new advances in the field of machine learning, data modeling and pattern recognition.
Research Objectives

Our objective is to utilize these relatively unexplored theoretical foundations in solving real world problem and make them worthy of clinical applications.
Our primary goals are to –

  • Establish links between medical measurements, e.g. brain function as measured by fMRI or brain structure as measured by MRI, and clinical/behavioral observations.
  • Model connectivity and causality networks in functional MRI
  • Do pattern recognition tasks pertaining to the brain images like estimating tissue density distributions and segmentation.

Approach and Methodology

We will begin by applying the recent theoretical approaches to applications in more or less their original form. Theoretical approaches we will be focusing on include constrained data decomposition and subspace learning, sparse feature selection, and graph based modeling of multivariate analysis. New approaches in machine learning are trying to address the problem of multivariate data decomposition using a part based approach rather than a holistic (global) ones like principal component analysis or vector quantization. This is achieved by putting constraints of non-negativity (Lee et al., 1999, 2001) and sparsity (Daubechies et al., 1998) on the decomposition. Graph embedding and extensions (Yan et al., 2007) is a promising technique that provides a generic formulation to unify all the different supervised and unsupervised algorithms within a common framework. New algorithms can be developed within this framework. The problem of regularization and variable selection especially in the case when number of predictors is much larger than the number of observations which is common in fMRI studies is of significant interest. Techniques like the lasso (Tibshirani, 1996) and elastic nets (Zou et al., 2004) are relatively new in this arena. Graph based modeling is a fresh and unconventional way of modeling multivariate data. Elegant techniques like small world networks (Bassett et al., 2006) which clusters data into dense local networks which are still connected to distant nodes via a small number of long range connections can be utilized as a theoretical basis to understand complex brain networks.

The ultimate goal is to adapt these algorithms to the chosen applications; utilizing a-priori knowledge of the field in general and the data under study in particular.


  1. Yan, S.C., Xu, D., Zhang, B.Y., Zhang, H.J., Yang, Q., Lin, S., 2007. Graph embedding and extensions: A general framework for dimensionality reduction. IEEE Transactions on Pattern Analysis and Machine Intelligence 29, 40-51.
  2. Bassett, D.S., Bullmore, E., 2006. Small-world brain networks. Neuroscientist 12, 512-523.
  3. Damoiseaux, J.S., Rombouts, S.A., Barkhof, F., Scheltens, P., Stam, C.J., Smith, S.M., Beckmann, C.F., 2006. Consistent resting-state networks across healthy subjects. Proc Natl Acad Sci U S A 103, 13848-13853.
  4. Lee, D.D., Seung, H.S., 1999. Learning the parts of objects by non-negative matrix factorization. Nature 401, 788-791.
  5. Zou, H., Hastie, T., 2005. Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society Series B-Statistical Methodology 67, 301-320.

Yeah I went for PhD 🙂

Print This Page Print This Page

Looking for something?

Use the form below to search the site:

Still not finding what you're looking for? Drop a comment on a post or contact us so we can take care of it!

Visit our friends!

A few highly recommended friends...