Clinical EEG In Modern Psychiatry: Diagnostics and Prediction

Introduction

Clinical electroencephalography is originally a neuropsychiatric technique which was developed and consistently further explored in the 1920s after extensive preliminary work, including animal experiments. Hans Berger was a psychiatrist and neurologist in Jena, Germany, where, after lengthy experiments which were challenging due to the technical conditions of the time, he was able to record electrical signals of the human brain from the surface of the head for the first time: signals that he originally referred to as ‘psychic energy’ (P.E.). In the course of this process, Hans Berger succeeded in detecting changes in cerebral electrical activity due to psychophysiological functional states (e.g., modulation of attention and vigilance, different states of wakefulness, opening and closing eyes) and mapping the influence of psychotropic substances.^1,2 In the 1930s, US pioneers such as Frederic and Erna Gibbs played a decisive role in introducing electroencephalography into clinical routine, recognizing its relevance not only in the field of epilepsy, but also in other neuropsychiatric disorders. ³ Despite competition from other examination methods - such as structural, functional or molecular imaging (CCT, MRI, fMRI, PET, SPECT) - and ongoing controversial discussions regarding specificity and sensitivity, electroencephalography has retained an important role in both clinical and research areas. It is a widely available, non-invasive and cost-effective method that can be used to directly assess brain electrical activity in vivo at a high temporal resolution (Table 1). As a functional procedure, EEG diagnostics play an indispensable role in psychiatry, in addition to the domain of epileptological disorders.^4-6

Furthermore, over the past decades, methodological advances in signal processing, quantitative EEG analysis and computational modelling have considerably expanded the analytical possibilities of electroencephalography. ⁷ These developments have enabled the extraction of reproducible electrophysiological features which extend beyond visual EEG interpretation and mere detection of abnormalities. In parallel, psychiatry has faced increasing pressure to improve treatment selection and reduce trial-and-error approaches, ⁸ particularly in major depressive disorder, where response rates to first-line interventions remain limited ⁹ despite a growing number of available treatments.

Importantly, this evolution does not imply a shift toward EEG-based diagnosis of psychiatric disorders. Psychiatric diagnoses remain fundamentally clinical constructs, defined by symptom patterns, course of illness and contextual factors. Consequently, an EEG marker that is “positive for depression” in an individual without clinical depression would be clinically meaningless. Instead, the emerging relevance of EEG in modern psychiatry lies in the domain of prediction rather than diagnosis. ¹⁰

Predictive EEG biomarkers aim to identify neurophysiological patterns which are associated with differential probabilities of treatment response, remission or adverse effects. These markers do not claim certainty at the individual level, but rather provide probabilistic information that can support clinical decision-making among equally indicated treatment options. ¹¹ Even when a prediction does not materialize in an individual patient, its clinical utility remains intact at the population level, where systematic shifts in response probabilities can translate into improved outcomes, reduced treatment duration and more efficient use of healthcare resources.

Against this background, the present paper integrates the classical clinical role of EEG in psychiatry with its emerging application as a source of predictive biomarkers, with a particular focus on depressive disorders. By distinguishing clearly between diagnostic limitations and predictive potential, we aim to provide a balanced and clinically grounded perspective on the current and future role of electroencephalography in psychiatric practice.

Methods

In clinical electroencephalography, cerebral cortical signals are recorded from the surface of the head with high amplification and noise reduction. The electrodes are positioned on the surface of the head according to the international convention (10/20 system). The signals result from excitatory and inhibitory postsynaptic sum activity of the upper cortical layers of the brain. The EEG initially reflects cortical brain activity, but it is also influenced by subcortical structures which are responsible for the rhythmization of EEG activity, among other things, and is modulated by neurochemical influences of the central neurotransmitter systems.^12,13 The normal electroencephalogram shows patterns and waveforms in characteristic frequency ranges which have traditionally been denoted by Greek letters dating back to the work of Hans Berger and are associated with different brain functional states (Table 2).

In clinical routine, the recording of the electroencephalogram is usually conducted in a relaxed state of wakefulness with closed eyes. Physiologically, an occipital alpha rhythm dominates which is usually easy to identify as a leading pattern (Figure 1). The alpha rhythm is regular and symmetrical (amplitudes usually slightly higher on the right side) and blocked by opening the eyes. This visual blocking reaction was already discovered by Hans Berger and is called the ‘Berger effect’.

In children and adolescents slow waves dominate while the typical alpha-EEG of adults develops during brain maturation. In the course of a longer recording the presence of the alpha rhythm can decrease with vigilance reduction (alpha attenuation). During fatigue and sleep theta and delta rhythms dominate in combination with graphoelements such as vertex potentials, sleep spindles or K-complexes that characterize different sleep stages. ¹² The use of hyperventilation, photostimulation or sleep deprivation, which are referred to as activation methods, can provoke and uncover pathological EEG patterns, which is why these procedures are particularly important in epilepsy diagnostics. ¹³ In patients with mental disorders, EEG recordings are usually well suited as non-invasive procedures, not least because of the direct personal contact with the examiner with the possibilities of verbal de-escalation and calming. ⁴ Nevertheless, an increased artifact load must be taken into account, especially in restless and agitated patients. Artifacts can make the analysis considerably more difficult and can lead to false positive evaluations; therefore, it is necessary to recognize and identify them beyond doubt. The following table shows a list of the most important biological and technical artifacts ¹² (Table 3).

Abnormal Patterns

Abnormal EEG patterns can be divided into diffuse (generalized) or circumscribed (focal) changes. Typical abnormal changes include suppression of the normal EEG rhythm, slowing of activity, increase in slow waves from the theta and delta frequency range (intermittent or continuous), periodic patterns, and the occurrence of epileptiform activity. If focal brain lesions are suspected, imaging diagnostics is the examination method of first choice and should be used for further diagnostic clarification of focal abnormal EEG findings. Electroencephalography as a brain functional examination is a complementary procedure which can provide initial indications of disorders even before the occurrence of imaging abnormalities. In metabolic-toxic disorders, but also in neurodegenerative diseases, commonly there is a slowing of the background activity as well as a generalized increase in slow theta and delta waves to different degrees, intermittently or continuously. Metabolic encephalopathies sometimes show characteristic patterns (triphasic waves) which are not specific but can provide valuable diagnostic clues. Periodic patterns (e.g., periodic sharp-wave complexes/PSWC with a frequency of 0.5 to 1.5 Hz) can indicate spongiform encephalopathies (Creutzfeldt-Jakob disease).^14,15 Epileptiform discharges are graphoelements that interrupt the background activity, break through it, show special configurations as sharp waves or spikes, and are to be assessed with regard to localization and field distribution. Epileptiform activity is considered to be an indicator of increased cerebral electrical excitability and is regularly found in patients with epilepsy.^12,13

Clinical Constellations with Abnormal EEG Patterns:

Clinical Use of Electroencephalography in Psychiatry

As a widely available and non-invasive brain functional diagnostic method, electroencephalography is particularly important for neuropsychiatric disorders that are not primarily associated with structural changes. One domain of the EEG remains the detection of epileptiform potentials or abnormal excitation patterns as an important diagnostic and differential diagnostic instrument in epileptology. However, neurophysiological examination procedures, and in particular classical electroencephalography, can also make an important contribution to psychiatry as a diagnostic and differential diagnostic tool: for diagnosis by exclusion, as an instrument for evaluating or monitoring the use of psychotropic drugs (neurotoxicity) and for the differential diagnosis of clinical conditions with alterations of consciousness and cognitive impairments. As a diagnostic screening instrument, the EEG can be used to validate the presence of metabolic-toxic disorders in delirious syndromes and contribute to the diagnostic classification.^4,5,17

Conceptual EEG Biomarker Framework: Diagnosis Versus Prediction

A central source of misunderstanding in the discussion of biomarkers in psychiatry arises from a failure to distinguish between diagnostic and predictive validity. Diagnostic biomarkers aim to identify the presence or absence of a disorder. In psychiatry this goal is inherently problematic. If an individual exhibits a so-called ‘biomarker for depression’ but does not meet clinical criteria for a depressive disorder, the biomarker is clinically meaningless. Conversely, a clinically depressed patient without the biomarker would be falsely classified. For diagnostic purposes such markers lack the necessary specificity and construct validity required to replace or supplement clinical assessment.

Prediction operates under a fundamentally different logic. Predictive biomarkers do not claim certainty at the individual level; instead, they modify probabilities. A predictive EEG marker can indicate that a given patient has a higher likelihood of responding to one treatment compared to another, without guaranteeing that response will occur. Crucially, the failure of a prediction in an individual patient does not invalidate the biomarker. Its utility emerges only when outcomes are aggregated across patient populations, where systematic probability shifts translate into improved overall response rates, shorter time to remission or reduced exposure to ineffective treatments. ³¹

Importantly, the EEG markers discussed here are derived from standard clinical resting-state recordings, typically obtained with eyes closed, and do not require task-based paradigms or event-related potential analyses, thereby facilitating integration into routine clinical workflows.

This distinction is well established in other areas of medicine, such as oncology or cardiology, but has been insufficiently emphasized in psychiatric biomarker research. EEG biomarkers in psychiatric disorders should therefore be evaluated not by their ability to ‘diagnose’ the entity, but by their capacity to enrich responder populations, guide treatment selection among equally indicated options and improve outcomes at the system level, thereby increasing the probability of faster and better response for individual patients.

Diagnostic EEG Biomarkers: Quantitative Limitations

For reasons of brevity, the following paragraphs will focus on EEG in major depressive disorder. Despite decades of research, the evidence does not support the use of EEG as a validated diagnostic tool for major depressive disorder. ³² Quantitative analyses have consistently shown a substantial overlap between EEG features observed in depressed patients and those seen in healthy individuals or patients with other psychiatric conditions. Factors such as vigilance regulation, medication effects, age, sex and comorbidity exert strong influences on EEG patterns, limiting diagnostic specificity.

Large-scale evidence syntheses demonstrate that certain EEG features can correlate with symptom severity or clinical subtypes, yet none achieve sufficient accuracy or robustness to justify a diagnostic claim. Notably, even when statistically significant group differences are observed, effect sizes are typically modest and unsuitable for individual-level classification. ³³ Consequently, EEG-derived markers can provide descriptive or pathophysiological insights, but they do not fulfill the criteria for diagnostic biomarkers in depression.

This conclusion is not a shortcoming of EEG technology per se but rather reflects the complexity and heterogeneity of depressive disorders. From a clinical perspective, the absence of diagnostic validity underscores the continued primacy of structured clinical assessment and reinforces the need to reframe EEG biomarkers within a predictive, rather than diagnostic paradigm. Further, the diagnostic value should be considered in light of the broader shift from categorical toward dimensional and transdiagnostic models in psychiatry. While EEG markers appear insufficiently specific to serve as diagnostic biomarkers for major depressive disorder, their sensitivity to physiological processes such as vigilance regulation, arousal, cognitive control, or affective reactivity may make them valuable for characterizing dimensions that cut across traditional diagnostic boundaries. In line with the RDoC framework, EEG may therefore contribute less to categorical diagnosis and more to biologically informed stratification of patients according to shared neurophysiological mechanisms. Empirical transdiagnostic RDoC-based work illustrates how such dimensional constructs can be examined across disorders rather than within single diagnostic entities alone.

Predictive EEG Biomarkers in Depression: Evidence and Scope

In contrast to diagnostic applications, a growing body of evidence supports the predictive use of EEG biomarkers in depression. ³⁴ Across multiple independent studies specific EEG features have been shown to differentiate responders from non-responders to particular treatments, with small-to-moderate effect sizes which are comparable to or exceed those of many established clinical predictors.^35-38 The following sections highlight selected EEG biomarkers that have shown the most consistent and clinically validated evidence for predicting treatment outcome in depression, and that are already being applied in clinical contexts. Importantly, the reported predictive markers extend across the full therapeutic spectrum, from first-line antidepressant treatments to second-line augmentation strategies, such as repetitive transcranial magnetic stimulation, and further to interventions used in treatment-resistant depression, including ketamine and electroconvulsive therapy (Figure 5).

Outlook: Predictive EEG Biomarkers Beyond Depression

While major depressive disorder currently represents the most extensively studied psychiatric condition with respect to EEG-based predictive biomarkers, there is no conceptual reason why such approaches should remain confined to depression. On the contrary, the neurophysiological principles underlying predictive EEG markers — such as cortical excitability, network synchronization, vigilance regulation and oscillatory dynamics — are transdiagnostic and relevant across multiple psychiatric and neuropsychiatric disorders.

In obsessive-compulsive disorder, preliminary evidence suggests that markers related to vigilance regulation, error monitoring and oscillatory activity in the theta and beta range may be associated with treatment response to selective serotonin reuptake inhibitors and cognitive-behavioral interventions.^46,47 In schizophrenia spectrum disorders, EEG markers reflecting excitation-inhibition balance, gamma-band synchronization and auditory evoked potentials have long been linked to pathophysiology, ⁴⁸ and EEG features are increasingly explored for predicting antipsychotic response or functional outcome. ⁴⁹

In dementia and mild cognitive impairment, there is increasing evidence that quantitative measures of EEG slowing, alpha peak frequency ⁵⁰ or connectivity are useful diagnostic tools. However, in terms of predicting disease progression or treatment responses, the evidence is still limited, mainly due to lacking prospective long-term data.

Importantly, the same conceptual caveats apply across these conditions as in depression. EEG biomarkers are unlikely to provide clear diagnostic classification with sufficient specificity, particularly in disorders characterized by clinical and biological heterogeneity. Their primary value lies instead more in prediction and stratification: identifying subgroups with differential probabilities of response, progression or adverse effects. As evidence accumulates, it is plausible that disorder-specific and transdiagnostic EEG biomarker profiles will emerge, enabling a more nuanced and biologically informed approach to psychiatric treatment selection across diagnostic categories.

Machine Learning and Deep Learning Approaches: Promise and Current Limitations

Recent years have seen a rapid increase in the application of machine learning, and in particular deep learning, to EEG data in psychiatry. These approaches offer theoretical advantages over conventional univariate or low-dimensional analyses, as they can capture complex, nonlinear interactions across channels, frequencies and time. In principle, deep neural networks can integrate spatial, temporal and spectral information in a way that more closely reflects the distributed nature of brain function. ⁵¹

However, despite promising proof-of-concept studies, several critical limitations currently prevent widespread clinical adoption. First, many machine learning studies in psychiatric EEG research are based on relatively small sample sizes, often in the low hundreds or fewer. In such settings complex models are prone to overfitting, leading to inflated performance estimates that fail to generalize to independent datasets. Second, methodological heterogeneity — including differences in preprocessing, feature extraction, model architecture and validation strategies — makes direct comparison across studies difficult and hampers cumulative knowledge building.

A central lesson from other fields of medicine is that clinical translation requires scale and replication. Large, multi-site studies with standardized protocols and independent validation cohorts are essential in order to establish robust predictive performance. Initiatives such as iSPOT-D or CAN-BIND illustrate the direction psychiatry must take, with sample sizes in the three- or even four-digit range, enabling reliable estimation of effect sizes and generalizability. Without such scale, even sophisticated deep learning models remain exploratory tools rather than clinically actionable solutions.

Equally important is transparency and interpretability. In clinical psychiatry, decision support systems must not only perform well statistically but also provide outputs that clinicians can understand, contextualize and critically evaluate. Black-box predictions without clear links to neurophysiological mechanisms risk undermining trust and adoption. As a result, hybrid approaches which combine biologically grounded EEG markers with machine learning-based integration may currently represent the most realistic path toward clinical utility.

From Research to Clinical Routine: Validation, Regulation and Acceptance

The transition of EEG biomarkers from research settings into routine psychiatric care requires more than scientific evidence alone. Validation and replication are necessary but not sufficient conditions. Clinical implementation must follow established pathways of certification and regulation, ensuring that biomarker-based decision support systems meet standards for safety, performance and intended use. This process is not a bureaucratic obstacle but a prerequisite for responsible clinical deployment.

Equally crucial is acceptance by clinicians. Psychiatrists must be convinced that decision support systems augment, rather than replace, clinical expertise. EEG-based biomarkers should be framed as additional sources of objective information that inform treatment selection among equally indicated options, not as automated decision-makers. Education, transparent communication of evidence levels and integration into existing clinical workflows will be key factors in achieving this acceptance.

Psychiatry can benefit from learning from other medical disciplines where decision support systems are already well established. In oncology, cardiology and intensive care medicine probabilistic models and risk scores are routinely used despite imperfect predictive accuracy, because their population-level benefits are well demonstrated. A similar cultural shift is required in psychiatry, moving from a focus on individual-level certainty to an appreciation of systematic improvement in outcomes.

Conclusion

Neurophysiological examinations – and in particular clinical electroencephalography – are important elements of additional brain functional diagnostics in neurology and psychiatry. ⁵² In the case of mental disorders rapid and reliable exclusion diagnostics are crucial for planning therapeutic strategies and initiating adequate treatment. Electroencephalography is complementary to imaging and contributes to differential diagnosis. Under pharmacotherapy, especially when using certain potentially neurotoxic drugs such as clozapine, lithium or valproate, the EEG is also suitable for follow-up examinations or monitoring.

Beyond these well-established applications, recent advances in quantitative analysis and computational methods have expanded the role of EEG in psychiatry. While current evidence does not support the use of EEG for diagnostic classification of psychiatric disorders, a growing body of research demonstrates that specific EEG-derived markers can predict treatment response and support therapeutic decision-making. Some of these markers are already used in the clinical context. Importantly, these predictive applications operate within a probabilistic framework and do not replace clinical judgment. Their clinical value emerges at the population level, where systematic shifts in response probabilities can lead to improved outcomes, reduced trial-and-error prescribing and a more efficient use of healthcare resources.

The further integration of EEG biomarkers into psychiatric routine care will depend on rigorous validation, replication in large and diverse patient cohorts, and adherence to regulatory standards. Equally, it will require a cultural shift toward the acceptance of decision support systems as complementary tools in clinical psychiatry. If implemented responsibly, EEG-based predictive biomarkers have the potential to enrich psychiatric practice while preserving its fundamentally clinical nature.