Dear List,
I have been following this long thread started by Matthew some time ago
and wanted to add my two cents. I am glad that the issue of inference in
neuroimaging has been brought back into focus and that so many had
interesting thoughts about it. I am also glad that after debate, a
fraction of the contributors has grouped around what I believe to be the
position so nicely expressed by Terry Jernigan and co-authors [Human
Brain Mapping 19:90 -95(2003)] which is.
1. However well designed the experiment may be, current statistical
methodology (thresholding) allows only inferences of association (point
A is associated with function F). This type of inference has brought the
field a long way forward but lack localization power as intended by
traditional neuropsychology
2. Although far from being optimal, non-thresholded maps are the only
available mean of attempting dissociation inference (function F is
associate more to point A than point B).
I understand the discomfort of some on issue number 2, but my take is
that we should not try to hide problems and limitations and instead take
issues like this as challenges, i.e. something to work on. I hope to see
a lot of new methodology on dissociation inference developed in the near
future.
BW
Federico
Federico E. Turkheimer, PhD. Senior Lecturer
Neuropathology Dept. Imperial College London
Charing Cross Campus St. Dunstan's Road
London, W6 8RP, UK Tel: +44 208 846 1174
Fax: +44 208 846 7794 Email:
[log in to unmask]
URL:
http://wwwfom.sk.med.ic.ac.uk/medicine/people/federico.turkheimer.html
|