Successful diagnosis of PTSD has been achieved using neural correlations from prewhitened magnetoencephalographic (MEG) time series (Georgopoulos et al. in J Neural Eng 7:16011, 2010. doi:10.1088/1741-2560/7/1/016011; James et al. 2015). Here, we show that highly successful classification of PTSD and control subjects can be obtained using neural correlations from prewhitened resting-state fMRI data. All but one PTSD (14/15; sensitivity = 93.3 %) and all but one control (20/21; specificity = 95.2 %) subjects were correctly classified using 15 out of 2701 possible correlations between 74 brain areas. In contrast, correlations of the same but non-prewhitened data yielded chance-level classifications. We conclude that, if properly processed, fMRI has the prospect of aiding significantly in PTSD diagnosis. Twenty-five brain areas were most prominently involved in correct subject classification, including areas from all cortical lobes and the left pallidum.
- fMRI classification