Determining of Predictors for Diagnostic Conclusion for Patients with Post-Traumatic Stress Disorder Using Class-Oriented Features Selection Technology and Group Method of Data Handling

This study explores the possibility of BOLD signals correlations usage obtained from resting-state functional magnetic resonance imaging (fMRI) of brain structures as predictors for the diagnostic assessment of patients with post-traumatic stress disorder (PTSD). To identify an informative ensemble of PTSD features, class-oriented feature selection technology was applied based on inter-class dispersion and the mRMR criterions. The use of logistic regression, optimized using the Group Method of Data Handling, for the classification of PTSD diagnoses allowed for the determination of the predictive forms of the identified informative characteristics. The resulting model demonstrated 96% accuracy, 92% sensitivity, and 100% specificity on the full dataset. The data was provided by the Institute of Nuclear Medicine and Diagnostic Radiology of the National Academy of Medical Sciences of Ukraine.