Combining neuroanatomical and clinical data to improve individualized early diagnosis of schizophrenia in subjects at high familial risk

Abstract

To date, there are no reliable markers for making an early diagnosis of schizophrenia before clinical diagnostic criteria are fully met. Neuroimaging and pattern classification techniques are promising tools towards predicting transition to schizophrenia. Here, we investigated the diagnostic performance of a combination of neuroanatomical and clinical data in predicting transition to schizophrenia in subjects at high familial risk (HR) for the disorder. Baseline structural magnetic resonance imaging (MRI) and clinical data from 17 HR subjects, who subsequently developed schizophrenia and an age and sex-matched group of 17 HR subjects who did not make a transition to the disease, yet had psychotic symptoms, were included in the analysis. We employed Support Vector Machine, along with a recursive feature selection technique to classify subjects at an individual level. Combination of both structural MRI and clinical data achieved an accuracy of 94% in predicting at baseline disease conversion in subjects at genetic HR. Overall, this paper presents a promising step in combining neuroanatomical and clinical information to improve early prediction of schizophrenia.