Application of Machine Learning Techniques for Distinguishing Schizophrenia Patients From Healthy Subjects Using Frontal Lobe Functions Assessments

Abstract

Machine learning (ML) represents a set of artificial in­telligence techniques that can assist in recognition of schizophrenia by classifying a person as belonging to either clinical or healthy subjects group. In the current study, we employed cognitive assessments of frontal lobe functions (the deficit of which is one of the prominent features of schizophrenia) for the training of ML models. Dataset for this research was engaged from our previous studies of two frontal lobe functions (response prepara­tion and inhibition of imitation) in case of schizophrenia. According to our knowledge, all previous cognitive ML schizophrenia studies used only the data from standard neuropsychological test batteries. Nevertheless, we em­ployed the cognitive data assessed with special experi­mental techniques that allowed us to engage Intra-indi­vidual reaction time variability (IIV) together with the classical reaction time (RT) assessment. It is important to emphasize that IIV is a cognitive measurement parameter that received vast attention of neuroscientists during the two last two decades and showed higher results in distin­guishing of schizophrenia patients from healthy subjects than standard RT in the number of studies. The result revealed statistically significant accuracy for all ML models in current study. Moreover, ML classifier with the highest accuracy outperformed the accuracy of a number of best models previously trained with standard neuropsychological test batteries datasets. Thus, cogni­tive experimental assessments of frontal lobe functions (response preparation and inhibition of imitation) can be effectively employed in developing of ML classifiers for distinguishing schizophrenia patients from healthy subjects.