Revisiting non-parametric activation detection on fMRI time series

In this paper, we propose some new ways of detecting activations in fMRI sequences that require a minimum of hypotheses and avoid any a priori modelling of the expected signal. In particular, we try to avoid linear assumptions and models. Instead, putting the emphasis on the dynamic evolution of the time series, we investigate its asymptotical behaviour. Considering an experimental block design, a key point is the ability of taking into account transitions between different signal levels, but still without the use of predefined impulse responses. The methods that we propose use well-known Student and information theoretical tests; they are based on the estimate of the asymptotic distribution of intensity values at a voxel when their time evolution is modelled as a first order Markov chain. The problem of statistical validation of these tests is also studied and a solution is proposed. The power of these methods seems high enough to avoid any smoothing, spatial or temporal, of the data. A first application is presented on a series of visual tasks obtained at Leuven University in order to characterise monkey motion perception. We compare our results with standard SPM maps.