Use predictive models and ensembles to statistically estimate similarities and differences between cohorts of brain networks.
Sparse Canonical Correlation Analysis (sCCA) is used as a bi-directional predictive model of brain connectomes (typically structural brain connectomes and functional brain connectomes). The sCCA biconvex criterion implemented in PMA R toolbox has been modified based on randomised Lasso principle to allow identification of the most relevant connections based on ensemble/bootstrapping principles. Therefore, relevant structural and functional connections that play an important role to the prediction are identified along with a probability score.
Identification.R projects functional connectivity matrices into an approximate tanget space on the Riemannian manifold, which allows to constrain prediction to Symmetric Positive Definite Matrices (SPD).
- F. Deligianni, H. Singh, H.N. Modi, S. Jahani, M. Yucel, A. Darzi, D.R. Leff, G.Z. Yang, 'Expertise and Task Pressure in fNIRS-based brain Connectomes', https://arxiv.org/abs/2001.00114
- F.Deligianni, J. Clayden and G.Z Yang, 'Comparison of Brain Networks Based on Predictive Models of Connectivity', IEEE BIBE, 2019. (Best Paper Award)
- F. Deligianni, D.W. Carmichael, Gary H. Zhang, C.A. Clark and J.D. Clayden, 'NODDI and tensor-based microstructural indices as predictors of functional connectivity', PLoS ONE, 11(4), 2016.
- F. Deligianni, M. Centeno, D.W. Carmichael and J.D. Clayden, 'Relating resting-state fMRI and EEG whole-brain connectomes across frequency bands', Frontiers in Neuroscience, 8(258), 2014.