Application of the functionally homogeneous regions (FHR) method to identify the most informative regions of the human brain for binary classification of schizophrenia based on resting-state functional MRI data

The article presents results of the analysis of the most informative brain regions for diagnosing schizophrenia based on resting-state functional MRI data using method of functionally homogeneous regions (FHR) previously developed by the authors and the CONN functional atlas. The analysis was performed using fMRI data from 32 subjects diagnosed with schizophrenia and 36 subjects from the control group obtained on Siemens tomograph. Data from 19 subjects diagnosed with schizophrenia and 29 subjects from the control group obtained on General Electric MRI scanner were used for verification. Eight most informative regions were identified. The analysis of the identified regions showed that changing the composition of the training group significantly affects the list of the most significant regions. At the same time the analysis of the identified most significant regions for repeatability with varying the composition of subjects showed that out of the eight identified most significant regions, four have repeatability higher than 70%, two have repeatability from 50% to 70%, and two have repeatability from 30% to 50%. This may indicate that the identified regions are not random and opens up prospects for further in-depth analysis and determination of their significance in diagnosing schizophrenia. Verification carried out on data from General Electric MRI scanner partially confirmed the heightened importance of the identified regions for the classification of schizophrenia pathology, but no perfect match was achieved on datasets from different MRI scanners.