Classification of first-episode schizophrenia spectrum disorders and controls from whole brain white matter fractional anisotropy using machine learning

Abstract

Background

Schizophrenia is a chronic disorder with an early onset and high disease burden in terms of life disability. Its early recognition may delay the resulting brain structural/functional alterations and improve treatment outcomes. Unlike conventional group-statistics, machine-learning techniques made it possible to classify patients and controls based on the disease patterns on an individual level. Diagnostic classification in first-episode schizophrenia to date was mostly performed on sMRI or fMRI data. DTI modalities have not gained comparable attention.

Methods

We performed the classification of 77 FES patients and 77 healthy controls matched by age and sex from fractional anisotropy data from using linear support-vector machine (SVM). We further analyzed the effect of medication and symptoms on the classification performance using standard statistical measures (t-test, linear regression) and machine learning (Kernel–Ridge regression).

Results

The SVM distinguished between patients and controls with significant accuracy of 62.34% (P = 0.005). There was no association between the classification performance and medication nor symptoms. Group level statistical analysis yielded brain-wide significant differences in FA.

Conclusion

The SVM in combination with brain white-matter fractional anisotropy might help differentiate FES from HC. The performance of our classification model was not associated with symptoms or medications and therefore reflects trait markers in the early course of the disease.

Disclosure of interest

The authors have not supplied their declaration of competing interest.