EEG-based BCI Systems in Neuropsychiatric Diseases

187

PD is the second most common neurodegenerative disorder, following

AD. PD is primarily identified by motor symptoms like hypokinesia, rest-

ing tremors, and a mask-like face. In addition to motor symptoms, non-motor

manifestations of PD encompass cognitive and sensory deficits, including hear-

ing loss, olfactory dysfunction, and sleep problems [81]. The utilization of con-

temporary AI techniques fed by various EEG characteristics plays a substan-

tial role in improving the sensitivity of clinical diagnosis of PD via cognitive

symptoms. This effect is observed even when symptoms are under control

with medication [82–85]. In the literature, both EEG in resting state [86] and

upon application of cognitive tasks were evaluated by using continuous data

or distinct extracted features such as Hjorth features [87] and phase locking

factor [88]. Moreover, some studies have aimed not only to distinguish patients

from healthy individuals but also to differentiate between different diseases.

For example, one of the latest studies [89] used a graph neural network fed by

effective brain connectivity to classify AD and PD (97.4% accuracy) as well

as PD and HC (94.2% accuracy).

SCZ is a serious mental health disorder characterized by disturbances in

thinking, emotions, and behavior, involving positive symptoms like hallucina-

tions and delusions, negative symptoms such as social withdrawal, and cogni-

tive impairments [90]. AI techniques with EEG features have also been em-

ployed for diagnosing SCZ. A recent comprehensive review by Jafari et al. [91]

summarized EEG-based ML and DL research for the diagnosis of SCZ with

various methodological approaches, and reported promising outcomes of the

studies, achieving maximum 100% performance.

Mood disorders are broadly divided into BD [92] and depressive disorders,

mainly including MDD [93]. While both bipolar disorder and major depressive

disorder involve episodes of depression, bipolar disorder is distinguished by the

occurrence of manic or hypomanic episodes. In the literature, both ML models

[94, 95] and DL models [96] were utilized to predict BD and depression. Also,

Yasin et al. [97] conducted a review of studies that adopted neural network

and deep learning approaches to detect both mental disorders.

7.5

Challenges and Future Perspectives

Although BCI technology gained momentum among researchers in the last

decades and numerous feature extraction and decoding techniques have been

implemented, achieving promising outcomes, several challenges in different as-

pects need to be overcome. These challenges could be methodological as well

as technological which consist of reliability, reactivity, and flexibility [17, 98].

In the study of Aggarwal and Chugh [45], multiple problems were categorized

based on ideal signal processing and classification methods, BCI function-

ing, performance assessment, and commercialization. Additionally, from the