Schizophrenia classification using machine learning on resting state EEG signal. (January 2023)

• Record Type: Journal Article
• Authors: Ruiz de Miras, J.; Ibáñez-Molina, A.J.; Soriano, M.F.; Iglesias-Parro, S.
• Is Part Of: Biomedical signal processing and control. Volume 79(2023)Part 2
• Publication Date: 2023-01
• Subjects: Schizophrenia -- EEG -- Resting state -- Machine-learning classification; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104233
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Early diagnosis of schizophrenia is a very difficult task. A processing pipeline for selecting features based on resting state EEG data is proposed. Complexity measures allowed standard machine learning algorithms to perform very efficiently. Abstract: Schizophrenia is a severe mental disorder associated with a wide spectrum of cognitive and neurophysiological dysfunctions. Early diagnosis is still difficult and based on the manifestation of the disorder. In this study, we have evaluated whether machine learning techniques can help in the diagnosis of schizophrenia, and proposed a processing pipeline in order to obtain machine learning classifiers of schizophrenia based on resting state EEG data. We have computed well-known linear and non-linear measures on sliding windows of the EEG data, selected those measures which better differentiate between patients and healthy controls, and combined them through principal component analysis. These components were finally used as features in five standard machine learning algorithms: k-nearest neighbours (kNN), logistic regression (LR), decision trees (DT), random forest (RF) and support vector machines (SVM). Complexity measures showed a high level of ability in differentiating schizophrenia patients from healthy controls. These differences between groups were mainly located in a delimited zone of the right brain hemisphere, corresponding to the opercular area and the temporal pole. Based on the area under the curve parameter in receiver operating characteristic curve analysis, we obtained high classification power in almost all of the machine learning algorithms tested: SVM (0.89), RF (0.87), LR (0.86), kNN (0.86) and DT (0.68). Our results suggest that the proposed processing pipeline on resting state EEG data is able to easily compute and select a set of features which allow standard machine learning algorithms to perform very efficiently in differentiating schizophrenia patients from healthy subjects.