Neural network analysis of the pattern of functional connectivity between cerebral areas in schizophrenia

Recent evidence suggests that the core abnormality of cerebral function in schizophrenia is a disruption of functional connectivity between diverse cerebral sites. Functional connectivity is defined as the correlation between neuronal activity at remote sites. It can be measured using functional imaging techniques such as positron emission tomography (PET). This paper reports an analysis using a neural network to discriminate between the patterns of functional connectivity in schizophrenic patients and healthy subjects. The data was derived from a PET study of regional cerebral blood flow during word generation in 6 healthy subjects and 16 schizophrenic patients with established illness, in whom the clinical diagnosis could be made with confidence. After training on data from two healthy subjects and seven schizophrenic patients, the neural network successfully assigned all members of a test set of four healthy subjects and nine schizophrenic patients to the correct diagnostic category. While this result should be interpreted with caution on account of the small sample size, it indicates that neural network analysis is potentially of value in the diagnosis of schizophrenia.