Predictive brain imaging of hallucinations

old_uid14603
titlePredictive brain imaging of hallucinations
start_date2014/11/07
schedule13h
onlineno
location_infosalle 01-02
detailsConférence de l’Axe IV. Hosted by Philippe Domenech
summaryHallucinations are complex and transient mental states associated with subtle and possibly brain-wide patterns of activity that are not reliably detected at the subject-level using conventional approaches. Linear support-vector machine classification (lSVM) can robustly learn individual spatial patterns of fMRI activity, while maintaining excellent generalization performances. Here, we used lSVM to build a decision function that accurately predicts whether an hallucination occur or not (H+/H-) during scanning and acquired fMRI data from schizophrenia patients (DSM-IV-TR) with frequent hallucinations to benchmark candidate classifiers. lSVM achieved a 71% accuracy in separating H+/H- periods, and improved to 0.81 sensitivity and 0.75 specificity (?2=4.9; p<.02) once trained to neglect speech imagery (normal inner speech). We expect to further improve these classification accuracies by taking into account more complex hallucinatory experiences and non-hallucinated resting-state data in SCZ patients.
responsiblesOliviero