The 5th Applied Artificial Intelligence and Robotics Laboratory Seminar showcased a presentation centered around the curation of the Nigerian Schizophrenia Dataset, known as NSzED, aimed at advancing computational psychiatry in Africa. The talk emphasized the ongoing EEG data acquisition process involving both schizophrenia patients and controls at the Obafemi Awolowo University Teaching Hospital and its subsidiary unit, Wesley Guild Hospital Unit, Ilesa. The presentation introduced the inaugural version of NSzED, effectively assessing the utility of the acquired EEG dataset by comparing EEG features in schizophrenia patients to those in healthy controls.
While numerous studies focus on detecting and classifying schizophrenia subtypes using EEG signals and computational methods, a majority of these studies use data acquired in developed regions, limiting the representativeness of the global schizophrenia populations. This hinders the development of generalizable models for diagnosing schizophrenia from EEG signals and adopting existing models in other developing and underdeveloped regions of the world. Therefore, intentional EEG data collection for computational schizophrenia investigation in developing and underdeveloped regions becomes crucial, highlighting the significance of datasets like NSzED, specifically collected to allow the investigation of EEG features reflective of or altered in schizophrenia.
The presentation unveiled the first version of NSzED, named NSzED-v1, containing multiple sessions of EEG recordings from 37 schizophrenia patients and 22 healthy controls, identified by the Mini International Schizophrenia Interview (MINI), World Health Organization Disability Schedule (WHODAS), and Positive and Negative Symptoms Scale (PANSS). The NSzED data acquisition protocol included four stages for each patient, incorporating auditory oddball stimuli task and fixed-frequency auditory stimuli to elicit mismatch negativity and auditory steady-state response, both consistent EEG biomarkers of schizophrenia.
Beyond data acquisition and organization, the presentation demonstrated efforts to verify the utility of the acquired EEG dataset by preprocessing the data and computing essential EEG features. Computed features included mismatch negativity amplitude, auditory steady-state response phase and amplitude, and fuzzy entropy values. Montage plots and statistical tests were utilized to compare these features between healthy controls and schizophrenia patients, revealing group-level differences. Furthermore, a basic neural network classifier was developed based on these features, achieving 89% accuracy and 86% F1-score, showcasing the dataset’s effectiveness in classifying schizophrenia and contributing to the diversity of existing EEG datasets for schizophrenia studies.
The seminar sparked thought-provoking discussions on Africa’s role in improving healthcare quality and the involvement of AI in the health sector. It also emphasized the crucial role of advanced technologies in unraveling psychiatric conditions such as schizophrenia.
