Reproducibility in group modeling

J.B. Poline
Commissariat à l'Énergie Atomique (CEA)

Functional MRI is a recent and powerful tool for studying the brain functions in both normal subjects and the patients. It has therefore inspired a wealth of interesting works in statistics, signal processing, image processing and modelling. There is however a gap between the current methods used in processing those data by the neuroscience or medical community and the most advanced methods proposed in the literature. Partly, this is due to the time needed for new technique to be understood and made available with the diffusion of software, and partly because the most interesting and difficult questions in the functional imaging community are only partially addressed. In this talk, I will focus on the problems involved in analysing the data from several subjects, in order to extract some knowledge on the parent population. The most used technique relies on a spatial normalisation of the subjects' brain and on standard statistics. I will point the limitations of this and describe new techniques improving both for spatial and the brain activity models aspect of the problem. The impact of those techniques on the reliability of group analyse will be described. I will then discuss what seem to be the future challenges in functional neuroimaging with a specific emphasis on databases.

Audio (MP3 File, Podcast Ready)

Back to Summer School: Mathematics in Brain Imaging