Patients with schizophrenia (SCZ) or bipolar disorder (BD) could be distinguished from healthy controls using dynamic functional connectivity data, according to results of a study published in the Journal of Neuroscience Methods.
This study was conducted at the Ankara University in Turkey. Patients with BD (n=30), SCZ (n=23), and healthy controls (n=31) underwent continuous wave Functional Near Infrared Spectroscopy evaluation while completing a mental state judgement task. Differences in functional connectivity between cohorts were used to discriminate between groups using 3 machine learning approaches.
The BD, SCZ, and control groups had a male:female ratio of 9:21, 19:4, and 23:8 (P =8.10´10-5); they were aged mean 39.9±13.1, 35.3±13.9, and 43.9±12.0 years; and they had completed 12.3±2.8, 12.0±2.7, and 14.9±3.0 years of education (P =.002), respectively. Among the patient cohorts, no significant difference in age at onset, hospitalization rate, or manic and depressive episodes were reported.
Among the SCZ and control cohorts, Dokuz Eylül Theory of Mind Index (DEToMI) scores negatively correlated with right Broca’s Area-left frontopolar area connection (r, -0.317; P =.038). Using connectivity data, the classification accuracy for SCZ compared with controls was 0.825 using support vector machine (SVM), 0.805 using discriminant analysis (DA), and 0.770 using K-nearest neighborhood (KNN) approaches. Sensitivity and specificity scores were 0.833 and 0.816 for SVM, 0.750 and 0.850 for DA, and 0.783 and 0.750 for KNN, respectively.
For BD compared with controls, DEToMI scores were correlated with the left middle temporal gyrus-right middle temporal gyrus connection (r, -0.376; P =.012). The accuracy, sensitivity, and specificity of discriminating between groups were 0.650, 0.616, and 0.700 for SVM; 0.790, 0.783, and 0.800 for DA; and 0.745, 0.700, and 0.783 for KNN, respectively.
Compared between patient groups, DEToMI scores negatively correlated with the left frontopolar area-left premotor cortex connection (r, -0.328; P =.029). SVM (accuracy, 0.755; sensitivity, 0.833; specificity, 0.666), DA (accuracy, 0.710; sensitivity, 0.800; specificity, 0.616), and KNN (accuracy, 0.710; sensitivity, 0.833; specificity, 0.583) approaches did not have sufficient specificities to distinguish between the 2 cohorts.
This study was limited by only considering cortical and not subcortical connections.
The study authors concluded, “In this study, we have shown that SCZ and BD groups can be discriminated from healthy controls subjects with acceptable sensitivity and specificity by machine learning approaches that uses mental state judgement based dynamic functional connections. The 2 patient groups also show different connections, but specificity values are not sufficient for differential diagnosis.”
Eken A, Akaslan DS, Baskak B, Münir K. Diagnostic classification of schizophrenia and bipolar disorder by using dynamic functional connectivity: an fNIRS study. J Neurosci Methods. 2022;376:109596. doi:10.1016/j.jneumeth.2022.109596