CPDD 72nd Annual Meeting • Scottsdale, Arizona - The College on ...

109
A WHOLE-BRAIN CLASSIFIER APPROACH FOR REAL-TIME
FMRI FEEDBACK TRAINING IN COCAINE ADDICTION.
A R Childress 1,2 , J Magland 1 , Z Wang 1 , A V Hole 1,2 , J J Suh 1,2 , A Fornash 1,2 , D
Willard 1 , R Fabianski 1 , R Carson 1 , R Hazan 1 , R N Ehrman 1,2 , T R Franklin 1 , M
Goldman 1 , R Szucs Reed 1 , C P O’Brien 1,2 ; 1 Psychiatry, University of
Pennsylvania School of Medicine, Philadelphia, PA, 2 VA VISN 4 MIRECC,
Philadelphia, PA
Aims: Region-specific real-time fMRI feedback has enabled subjects to control
pain-modulating brain activity, and to achieve pain relief. <strong>The</strong>se pioneering studies
encourage similar rtfMRI feedback attempts for “craving circuit modulation”
in addiction, but the relevant brain regions are not yet fully characterized, may
be spatially distributed, and may vary across individuals. We are thus testing
whether a “whole brain” classifier approach may be used 1) to rapidly distinguish
the brain states associated with viewing cocaine vs. non-drug videos, and 2)to
provide rtfMRI feedback for controlling the brain state associated with viewing
a cocaine video.
Methods: BOLD fMRI at 3T with a partial least squares (PLS)linear classifier
was used to characterize the whole brain response to alternating 30 sec cocaine
vs. non-drug videos in cocaine patients (n=5; ongoing) and controls (n=2; ongoing).
Real-time visual feedback based on this classifier was provided to two controls
via a cursor adjacent to the ongoing video, with instructions to attempt control
of the cursor by increasing or decreasing the state associated with viewing the
target video.
Results: <strong>The</strong> whole-brain classifier based on PLS regression robustly distinguished
(15