Classification of schizophrenic patients and healthy controls using [18F] fluorodopa PET imaging.

Striatal dopaminergic overactivity has been implicated in the pathophysiology of schizophrenia on the basis of in vivo neuroimaging studies. In particular, elevated striatal dopamine synthesis and storage has been repeatedly demonstrated in schizophrenia using the radiotracer 6-[18F] fluoro-l-DOPA ([18F] DOPA) and positron emission tomography (PET). Conventionally analysed [18F] DOPA PET imaging lacks the sensitivity or specificity to be used diagnostically. The aim of this study was to determine if the application of an Artificial Neural Network (ANN) would improve classification of images, and increase the sensitivity and specificity of [18F] DOPA as a potential diagnostic test for schizophrenia. We tested an ANN model in the discrimination of schizophrenic patients from normal controls using [18F] DOPA rate constants within the anterior-posterior subdivisions of the striatum, and compared the model with a general linear analysis of the same data. Participating in the study were 19 patients diagnosed with paranoid schizophrenia and 31 healthy subjects. Maximum classification was achieved using laterality quotients, - the ANN model correctly identified 94% of the controls and 89% of the patients, equivalent to 89% sensitivity and 94% specificity. Using all bilateral striatal regions correctly categorised 74% of the controls and 84% of the patients, equivalent to 84% sensitivity and 74% specificity. In comparison, the general linear analysis performed poorly, correctly classifying only 58% of the controls and 63% of the patients. Overall, these analyses have shown the potential utility of pattern recognition tools in the classification of psychiatric patients based upon molecular imaging of a single target.