Smart prediction models focused on neuroimaging in psychiatric diagnosis and classification

Research designed and conducted by the Department of Psychiatry, Mental Health and Hospital, Üsküdar University Faculty of Medicine, shared the effective results of machine learning methods in the diagnosis and classification of psychiatric diseases. The study revealed the distinctive biomarker potential of EEG in diagnosing various psychiatric disorders.

In the research conducted by the Department of Psychiatry, Mental Health and Hospital, Üsküdar University Faculty of Medicine, EEG data from individuals with bipolar disorder, attention deficit hyperactivity disorder (ADHD), depression, obsessive-compulsive disorder (OCD), opioid addiction, post-traumatic stress disorder (PTSD), schizophrenia, and healthy individuals were used as biomarkers, and a prediction model to support diagnosis was developed using computer-aided machine learning methods. 

Can psychiatric disorders be diagnosed and classified with EEG data?

The findings showed that by using EEG data as biomarkers, it is possible to predict with a high degree of accuracy whether a client has a psychiatric disorder. Accordingly, after a person consulting a doctor with a complaint is generally evaluated, an accurate diagnosis of psychiatric diseases can be made with EEG recording, distinguishing them from other disease groups with their distinctive features. 

Prof. Dr. Nevzat Tarhan: "Machine learning will contribute to the field of psychiatry"

Summarizing the research, Üsküdar University Founding Rector, Psychiatrist Prof. Dr. Nevzat Tarhan said, “While trying to differentiate between numerous and various disease categories related to the subject, we can say that some diseases (ADHD, depression, schizophrenia) can be better distinguished through model-based analysis. Considering the findings, it is predicted that the analyses obtained from this study will contribute to future research in psychiatry using machine learning. Especially prediction models developed with new-generation in-silico methods yield significant clinical results in this field. Although old-generation superficial learning algorithms and statistical models have contributed to the literature, new-generation learning algorithms, especially those based on deep learning capable of extracting features, contribute to the development of studies in neuroscience, diagnostic and prognostic processes, early diagnosis, and accurate treatment processes for clinicians.” 

Prof. Dr. Nevzat Tarhan stated, “In the diagnosis of psychiatric disorders, physicians follow a symptom-based approach. According to this approach, internationally valid process-based diagnostic tools like the Diagnostic and Statistical Manual of Mental Disorders (DSM) or the International Classification of Diseases (ICD), along with patient reports, observation, and the physician’s experience, are utilized. As in other areas of medicine, the search for biomarkers used in disease-related processes continues in psychiatry.” 

The study, which also included Prof. Dr. Cumhur Taş, a faculty member of the Department of Mental Health and Diseases, Üsküdar University Faculty of Medicine, was published in the ‘International Journal of Medical Informatics’.  Within the scope of the study, a dataset containing electroencephalogram (EEG) measurements of patients (550 patients) diagnosed with various psychiatric diseases was analyzed using machine learning methods, and the diseases were classified with the models obtained.

About the Research: DOI    10.1016/j.ijmedinf.2022.104926

Research Link: http://dx.doi.org/10.1016/j.ijmedinf.2022.104926