Crossing-fibres and TBSS
In [Jbabdi 2010] we presented an approach for feeding in crossing-fibre partial volume fractions into TBSS, including a method for trying to make sure that the same tract (whether primary or secondary) is compared across all subjects. The script for using crossing fibre volume fractions in tbss is called tbss_x, and its usage is described in the main documentation. The script uses two commands, swap_voxelwise and swap_subjectwise, which can be run independently of tbss_x in order to ensure cross-subject correspondence of crossing-fibre values.
Nonstationarity and TBSS
In [Salimi-Khorshidi 2010] we addressed the general question of nonstationarity for permutation testing. We hope that this can be used in TBSS to reduce causes of nonstationarity when testing on the skeleton, for example, caused by: varying spatial smoothness, varying cross-subject variance, and varying neighbourhood size on the skeleton (for example, when there are differently-sized disconnected skeleton chunks, or where the size of the local neighbourhood depends on local skeleton orientation). The two-stage correction process can be applied in randomise by using the --twopass option.
TBSS and multivariate testing
In [Groves 2010] we presented work on multivariate data-driven analysis across multiple modalities, including TBSS-preprocessed data (in this context, for example, FA and MD are "different" modalities). This both takes advantage of modelling across multiple voxels and multiple modalities in a single ICA-based analysis. The Bayesian framework automatically adjusts aspects of the modelling such as the relative weighting of the different modalities, and estimation of the number of distinct components that the data supports. Single-modality TBSS-preprocessed data (e.g. just the FA data in all_FA_skeletonised) can already be fed into melodic, while the multi-modal approach being developed by Groves et al is still in development, and will hopefully be released as part of FSL in the future.