Diagnosis of Schizophrenia and Its Subtypes Using MRI and Machine Learning.

PURPOSE

The neurobiological heterogeneity present in schizophrenia remains poorly understood. This likely contributes to the limited success of existing treatments and the observed variability in treatment responses. Our objective was to employ magnetic resonance imaging (MRI) and machine learning (ML) algorithms to improve the classification of schizophrenia and its subtypes.

METHOD

We utilized a public dataset provided by the UCLA (University of California, Los Angeles) Consortium for Neuropsychiatric Research, containing structural MRI and resting-state fMRI (rsfMRI) data. We integrated all individuals within the dataset diagnosed with schizophrenia (N = 50), along with age- and gender-matched healthy individuals (N = 50). We extracted volumetrics of 66 subcortical and thickness of 72 cortical regions. Additionally, we obtained four graph-based measures for 116 intracranial regions from rsfMRI data, including degree, betweenness centrality, participation coefficient, and local efficiency. Employing conventional ML methods, we sought to distinguish the patients with schizophrenia from healthy individuals. Furthermore, we applied the methods for discriminating subtypes of schizophrenia. To streamline the feature set, various feature selection techniques were applied. Moreover, a validation phase involved employing the model on a dataset domestically acquired using the same imaging assessments (N = 13). Finally, we explored the correlation between neuroimaging features and behavioral assessments.

FINDING

The classification accuracy reached as high as 79% in distinguishing schizophrenia patients from healthy in the UCLA dataset. This result was achieved by the k-nearest neighbor algorithm, utilizing 12 brain neuroimaging features, selected by the feature selection method of minimum redundancy maximum relevance (MRMR). The model demonstrated effectiveness (72% accuracy) in estimating the patient's label for a new dataset acquired domestically. Using a linear support vector machine (SVM) on 62 features obtained from MRMR, patients with schizophrenic subtypes were classified with an accuracy of 64%. The highest Spearman correlation coefficient between the neuroimaging features and behavioral assessments was observed between the degree of the postcentral gyrus and mean reaction time in the verbal capacity task (r = 0.49, p = 0.001).

CONCLUSION

The findings of this study underscore the utility of MRI and ML algorithms in enhancing the diagnostic process for schizophrenia. Furthermore, these methods hold promise for detecting both brain-related abnormalities and cognitive impairments associated with this disorder.