Depression is a mental disorder that involves emotional symptoms, such as depressed mood, and encompasses a variety of physical and cognitive changes, including alterations in appetite, sleep patterns, and concentration [17]. Factors that contribute to the onset of depression include genetic susceptibility, stress, and developmental trauma. However, the lack of objective biomarkers often makes it difficult to receive timely and appropriate treatment [18]. Additionally, even with pharmacological treatment, 30% to 40% of patients experience residual symptoms, making it crucial to predict treatment response; however, there are currently no adequate biomarkers [19]. Therefore, ongoing efforts have been made to use biomarkers for the diagnosis and treatment of depression, and various studies have been conducted using EEG.

In the context of using EEG as a biomarker for depression, characteristics based on Fourier transform spectral analysis have been the most extensively studied. Alpha asymmetry is a representative marker of depression based on spectral analysis, reflecting hypoactivity of alpha power in the left frontal area and hyperactivity in the right frontal area [20]. However, recent studies have suggested that these characteristics are not consistently observed and remain controversial [21]. More recently, increased gamma-band power, located in higher frequency ranges, has been identified as a novel biomarker of depression, indicating changes in the balance between brain excitation and inhibition [22]. EEG during sleep shows specific changes in individuals with depression. Specifically, patients with depression exhibit increased rapid eye movement (REM) sleep and decreased REM sleep latency. Additionally, changes in slow-wave activity related to treatment response have been identified [23]. REM sleep is characterized by changes in the theta-, beta-, and gamma-band activities [24]. Considering these characteristics, various methods based on Fourier transformation have been utilized for the automatic classification of sleep stages, including REM and slow-wave sleep [25]. Although Fourier transformation primarily reflects the linear characteristics of EEG signals, complexity measures that capture the nonlinear characteristics of EEG signals have been used in various analyses of depression [26]. Studies using various complexity measures, such as entropy and Lempel–Ziv complexity, have generally indicated higher complexity in the EEG of individuals with depression [27,28]. Lempel–Ziv complexity is a metric that reflects the complexity of time-series data, whereas entropy, based on information theory, reflects the uncertainty of time-series data. In EEG, these metrics reflect fluctuations in neural activity or spontaneous stochasticity [29-31]. Studies using event-related potentials (ERPs) have been conducted. The P300 component, which reflects cognitive processing in the brain, tends to show an increased latency and decreased amplitude during depression [32]. Furthermore, the P50 component, which reflects sensory processing, shows increased amplitude in depression, indicating impaired sensory gating in patients [7].

BD is a chronic mental illness characterized by extreme mood swings between depression and mania, affecting approximately 2.4% of the global population and primarily manifesting in early adulthood [33]. The development of reliable biomarkers is essential for the accurate diagnosis and prediction of treatment outcomes in BD and for elucidating its uncertain pathophysiology [34]. EEG provides detailed insights into brain activity through the analysis of various brainwave frequencies such as alpha, delta, theta, and gamma [35]. In patients with BD, EEG studies have identified several notable abnormalities. For example, abnormalities in alpha activity, particularly in frontocentral regions, have been associated with cognitive and sensory processing deficits [36,37]. While beta, left delta, and theta activities are normalized with lithium, treatment response is most closely associated with basal delta activity. Lithium plasma concentration correlates with theta activity [38], and delta activity is linked to attention deficit and cognitive impairment [39]. Moreover, the increased beta activity and frontal alpha asymmetry during manic episodes suggest neural excitation and mood dysregulation [40]. Similar to research on depression, studies have been conducted to identify biomarkers of BD using measures of complexity. Fernández et al. [41] used Lempel–Ziv complexity to distinguish the differences in EEG complexity between depression and BD. Additionally, Bahrami et al. [42] reported that EEG complexity increased during the manic state in BD.

Schizophrenia is a persistent and often debilitating mental illness that affects approximately 1% of the global population [43]. Developing biomarkers for schizophrenia is essential to provide objective diagnostic and treatment measures, validate targets, predict responses, and enable personalized treatment by identifying disease mechanisms [44]. In schizophrenia research, power spectral analysis of EEG has revealed several key abnormalities. Patients typically exhibit increased power in the delta (0.5–4 Hz) and theta (4–8 Hz) frequency bands, indicating a disruption in slow-wave activity [45]. Additionally, reduced alpha power (8–13 Hz) is commonly observed, reflecting impaired cortical inhibitory mechanisms and resting-state neural activity. Although alterations in beta power (13–30 Hz) have been reported, the findings vary, with a few studies showing increased beta activity and others showing decreased beta activity.

Consequently, various EEG-based markers have been extensively studied in schizophrenia, with a particular focus on ERPs that measure electrophysiological changes [45] following stimulation. The most studied markers are P50 sensory gating, N100, mismatch negativity (MMN), and P300, each of which provides unique insights into different aspects of brain function. P50 sensory gating reflects the ability of the brain to suppress irrelevant stimuli, with deficits indicating impaired inhibitory processes that are commonly observed in schizophrenia [46]. The amplitude of the N100 waveform, which is associated with early auditory processing, is typically reduced in patients with schizophrenia, suggesting deficiencies in sensory registration [47]. MMN, elicited by deviations in the sequence of auditory stimuli, is a robust marker of predictive coding deficits and is sensitive to N-methyl-ᴅ-aspartate receptor dysfunction, highlighting its potential as an early indicator of psychosis risk [48]. Finally, the P300 component, which is indicative of attention and memory processes, shows consistent amplitude reduction and latency prolongation in schizophrenia, correlating with cognitive impairment [49]. These EEG markers enhance our understanding of the underlying neurobiology of psychosis and offer valuable tools for early diagnosis, monitoring disease progression, and tailoring individualized treatment strategies.

EEG-based biomarkers offer several advantages for neuropsychiatric applications [50]. They enable real-time data acquisition and provide immediate feedback and continuous monitoring, which are crucial for rapid diagnosis and treatment response assessment [51]. In addition, unlike other neuroimaging techniques that indirectly measure neuronal activity through blood oxygen levels, EEG directly measures the electrical activity between neurons, which is the primary cause of psychiatric disorders [52]. EEG is noninvasive and causes minimal discomfort to patients, making it ideal for repeated measurements [53]. The noninvasive nature of EEG makes it significantly safer than other invasive methods used to diagnose neuropsychiatric disorders, such as blood sampling, cerebrospinal fluid tapping, and brain tissue biopsy. Moreover, EEG is cost-effective compared to other neuroimaging techniques, facilitating broader patient access and large-scale studies [54]. With high temporal resolution, EEG captures changes in fine-grained brain activity that are essential for understanding the mechanisms of neuropsychiatric diseases [55]. EEG is applicable for diagnosing and monitoring various conditions, including epilepsy, sleep disorders, ADHD, and schizophrenia, and can predict treatment responses and disease progression [11,40]. Integrating EEG data with machine learning algorithms allows the development of precise diagnostic models, aiding in the creation of personalized treatment plans [2,56]. Furthermore, with appropriate data processing and analysis techniques, EEG-based biomarkers can achieve high sensitivity and specificity, making them useful for early diagnosis and evaluation of treatment efficacy [11]. These advantages highlight the significant role of EEG-based biomarkers in the diagnosis and management of neuropsychiatric disorders.

Despite their numerous advantages, EEG-based biomarkers for psychiatric disorders have several limitations that restrict their broader clinical application. First, the spatial resolution of EEG is relatively low compared with that of other neuroimaging techniques such as functional magnetic resonance imaging (fMRI), making it difficult to accurately localize neural activity, particularly in deeper or smaller brain regions [57]. This limitation can impede the precise mapping of brain functions and identification of specific neural substrates underlying neuropsychiatric conditions. Second, EEG signals are highly susceptible to interference [58]. EEG signals can be distorted by the electrical resistance of the skull and scalp and are often contaminated by external electromagnetic interference and physiological noise, such as eye blinks, muscle movements, and cardiac activity. This interference can significantly affect the quality and reliability of the recorded data, making it challenging to isolate relevant neural signals from noise. Therefore, distinct changes in EEG, such as those observed with seizures, can be detected with minimal processing and near real-time monitoring. However, the cognitive and emotional characteristics of the brain, which are primarily associated with psychiatric disorders, are highly susceptible to such noise [59]. Third, the complexity and inherent noise of EEG data make their interpretation particularly challenging [60].

Analyzing EEG data requires sophisticated signal-processing techniques and expert knowledge of neurophysiology and bioinformatics, which can be significant barriers in both clinical and research settings. However, the need for advanced analytical methods limits the accessibility and practicality of EEG-based biomarkers, particularly in settings with limited technical resources and expertise. Additionally, individual physiological differences, such as variations in skull thickness, scalp conditions, and brain anatomy, can markedly influence EEG signals [61]. These interindividual differences can lead to variability in the recorded data, thus reducing the consistency and reliability of EEG-based biomarkers. Such variability poses a significant challenge in standardizing EEG measurements and ensuring that biomarkers are comparable across individuals and studies. Another critical limitation is the difficulty in standardizing EEG measurements and data processing methods [62]. Considerable variability exists in the equipment used, protocols for data acquisition, and algorithms for data analysis, which can lead to discrepancies in the results across different studies [63]. For example, high-density EEG systems, which provide more detailed spatial information, may yield different results than low-density systems, complicating the comparison and replication of findings. Finally, the clinical applicability of EEG-based biomarkers is currently limited to only a few neuropsychiatric conditions. Although significant progress has been made regarding the use of EEG biomarkers for disorders such as epilepsy, schizophrenia, and sleep disorders, the validity and reliability of these biomarkers need to be established for a broader range of conditions through extensive research and clinical validation [5]. This limitation underscores the need for further studies to explore the potential of EEG biomarkers in the diagnosis and management of other neuropsychiatric and neurological disorders. These challenges highlight the need for continued technological advancements, efforts to standardize EEG methodologies, and comprehensive research to enhance the reliability, validity, and clinical applicability of EEG-based biomarkers. Addressing these limitations is crucial for integrating EEG biomarkers into personalized medicine and for improving patient outcomes in neuropsychiatric care.

Developing EEG-based biomarkers for psychiatric disorders presents significant challenges owing to both the limitations of EEG as a modality and the inherent complexity and heterogeneity of these conditions. Unlike other medical fields, where biomarkers (such as genetic markers in Huntington’s disease or breast cancer gene 1/2 mutations) provide clear diagnostic value, psychiatric disorders lack such straightforward biological markers [5]. The symptoms of psychiatric disorders are often nonspecific and overlap significantly across diagnoses, making it difficult to identify unique biomarkers [64]. For instance, symptoms of psychosis can manifest in schizophrenia, BD, major depressive disorder with psychotic features, or even because of substance abuse.

The diagnostic criteria in psychiatry, such as those outlined in the Diagnostic and Statistical Manual of Mental Disorders, fifth edition, are primarily based on symptomatology rather than underlying biological mechanisms, further complicating the identification of reliable biomarkers [65]. This symptom-based approach leads to diagnostic instability, whereby a patient’s diagnosis may change over time as the symptoms evolve, potentially leading to misclassification in biomarker studies. Moreover, the lack of objective laboratory tests in psychiatry means that clinical diagnoses rely heavily on subjective clinical interviews, increasing variability and reducing the reliability of diagnoses among clinicians. This issue is compounded by the high comorbidities of psychiatric conditions, in which patients often present with multiple overlapping disorders, complicating the isolation of biomarkers specific to a single diagnosis [2]. Methodological challenges, such as small sample sizes, lack of standardization of study protocols, and need for longitudinal studies to understand the temporal relationships between biomarkers and psychiatric conditions, further hinder progress [5]. Many studies have excluded individuals with severe symptoms or high comorbidities, limiting the generalizability of their findings to broader clinical populations.

Animal models, which are instrumental in biomarker research for other medical conditions, are less effective in psychiatry because of the fundamental differences in brain structure and function between humans and animals [66]. Psychiatric disorders often involve complex cognitive and emotional processes that are difficult to replicate in animal models, thereby reducing their utility for biomarker discovery. These challenges highlight the need for large-scale collaborative research efforts that integrate multiple data types and adopt a more systematic and standardized approach to biomarker research. By addressing these methodological and clinical challenges, the field of psychiatry can move closer to identifying reliable biomarkers to improve the diagnosis, treatment, and understanding of psychiatric disorders.