Early identification of Schizophrenia using multi-view learning model

• Published in: Cogent engineering Vol. 11; no. 1
• Main Authors: Y, Krishna Bhargavi, Ritwika, Singamsetty, Lakshmi Prasanna, Javadi, Prashanthi, Bhupathi, Sankara Babu, Bulipe, Nemova, Darya Viktorovna, Joshi, Abhishek, Mohammed, Al-Farouni
• Format: Journal Article
• Language: English
• Published: Taylor & Francis Group 31.12.2024
• Subjects: actigraph; EEG; EHR; FMRI; multi-view learning model; Schizophrenia
• ISSN: 2331-1916; 2331-1916
• DOI: 10.1080/23311916.2024.2384649

The modern lifestyle and work culture are impelling the human beings into various health issues, both physically and mentally. Schizophrenia is one such prevalent mental illness that most adults are facing these days. The most common symptoms that Schizophrenic patients exhibit are classified into positive and negative symptoms. The positive symptoms include hallucinations, abnormal movement, delusions and disorganized thinking. The negative symptoms involve Anhedonia, Alogia, Asociality, Avolition, etc. The patients’ exhibit assorted such symptoms and involves several such features namely cognition, language, emotion, interest levels and motor inhibitions. It is a psychiatric disorder of brain, and the traditional machine learning models are unable to handle the high dimensional non-linear nature of the brain and understand Schizophrenia to the full extent. The existing methods like Hopfield Networks, CNN, ANN and RNN are implemented to identify Schizophrenia by considering only one of the aforementioned symptoms. Multi-view learning model is employed in this article for early prognosis of Schizophrenia taking into consideration multiple aspects like Electroencephalogram (EEG) pursuing the electrical activity records of brain, Functional Magnetic Resonance Imaging (fMRI) measuring the change in blood flow in response to the activity of brain, Electronic Health Record (EHR) that captures the digitized patient’s health record, Actigraph data that records the motor inhibitions from wearable sensor devices. It is identified that the multi-view learning model outperformed the existing methods when compared against Accuracy, Recall Precision and F1 Score. The proposed method obtained these metrics with values 93.6%, 93.9%, 88.7% and 90.8% respectively.