Classification Models for Diagnosing Schizophrenia With MRI Neuroimaging
A classification model incorporating multimodal brain features, including gray and white matter abnormalities, was effective in identifying schizophrenia in patients.
A classification model incorporating multimodal brain features, including gray and white matter abnormalities, was effective in identifying schizophrenia in patients, according to data published in Schizophrenia Bulletin. These results may have implications for the current diagnostic techniques used to identify schizophrenia spectrum disorder.
Researchers acquired structural magnetic resonance imaging (sMRI) and diffusing tensor imaging (DTI) data from 152 patients with first-episode schizophrenia and 154 healthy control participants. A Gradient Boosting Decision Tree was used to determine the brain measures most strongly associated with schizophrenia, including "regional gray matter volume, cortical thickness, gyrification, fractional anisotropy, and mean diffusivity." The classification model was developed and validated using the imaging data from the first 98 patients with schizophrenia and 106 matched control participants (Dataset 1), then tested using independent data from the final 54 patients with schizophrenia and 48 control participants (Dataset 2).
In developing the classification model, the brain features most frequently associated with schizophrenia were "cortical thickness of left transverse temporal gyrus and right parahippocampal gyrus." Additionally, the fractional anisotropy of the "left corticospinal tract and right external capsule" had high discriminative power. These features combined had a classification accuracy of 75.05% for Dataset 1. An average accuracy of 76.54% in Dataset 2 was achieved analyzing features of cortical thickness, gyrification, fractional anisotropy, and mean diffusivity. Each of the studied brain modalities could be classified as "gray matter abnormalities" or "white matter disruptions" and may be significant in the pathology of schizophrenia.
These data underscore the potential use of "multimodal data fusion" in the diagnosis of schizophrenia. Further studies to identify additional schizophrenic brain modalities is necessary to improve the efficacy of neuroimaging diagnostic techniques.
Liang S, Li Y, Zhang Z, et al. Classification of first-episode schizophrenia using multimodal brain features: a combined structural and diffusion imaging study [published online June 27, 2018]. Schizophr Bull. doi: 10.1093/schbul/sby091