Schizophrenia in the age of artificial intelligence: A review of advances in diagnosis, prediction, and digital psychiatry

Abstract

Schizophrenia is a chronic and disabling psychiatric disorder affecting approximately one percent of the world’s population. It manifests through positive, negative, and cognitive symptoms that severely impair social and occupational functioning. Despite extensive research, diagnosis remains primarily subjective, based on Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition criteria, and effective early intervention is still limited. This narrative review synthesizes current evidence from epidemiological, neurobiological, and clinical studies alongside recent advances integrating artificial intelligence (AI) into schizophrenia research. Literature sources were drawn from PubMed, Scopus, and Web of Science, focusing on studies addressing etiology, neuroimaging findings, treatment outcomes, and AI-based diagnostic approaches. The etiology of schizophrenia is multifactorial, involving genetic vulnerability, neurodevelopmental disturbances, neurotransmitter dysregulation, and environmental stressors such as perinatal complications and substance use. Neuroimaging findings consistently reveal gray matter reduction, ventricular enlargement, and prefrontal-temporal connectivity abnormalities. Pharmacological management especially with second-generation antipsychotics such as clozapine, risperidone, and olanzapine remains the treatment cornerstone, supported by psychosocial interventions that improve adherence and functional recovery. Emerging AI-driven tools using neuroimaging, electroencephalography, and behavioral data show high diagnostic accuracy and potential for personalized treatment planning. Schizophrenia continues to present diagnostic and therapeutic challenges due to its biological complexity and clinical heterogeneity. The review stands out because it uses modern AI methods to study schizophrenia for better diagnosis and treatment planning and earlier disorder identification. The review combines traditional medical methods with new computational systems to demonstrate schizophrenia research progress while developing vital ethical standards and procedural systems for psychiatric care during the AI age.

Key Words: Schizophrenia; Artificial intelligence; Explainable artificial intelligence; Electroencephalography; Magnetic resonance imaging; Neuroimaging; Speech analysis; Digital psychiatry; Machine learning; Precision psychiatry

Core Tip: Schizophrenia remains one of the most complex psychiatric disorders, characterized by profound neurobiological, cognitive, and social dysfunctions. Despite decades of clinical and neuroimaging research, diagnosis still depends largely on subjective observation. This review integrates classical perspectives on epidemiology, etiology, and treatment with recent developments in artificial intelligence (AI). It highlights how AI-driven models-particularly those analyzing neuroimaging and electroencephalography data-offer objective insights into brain alterations and symptom mechanisms. By bridging traditional psychiatry with emerging computational tools, the paper outlines a roadmap for early detection, personalized intervention, and transparent decision support in schizophrenia care.

INTRODUCTION

Schizophrenia is a complex neurobiological disorder that affects millions of people worldwide and has profound consequences for lifelong functioning. The Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) defines schizophrenia through its criteria which produce a broad range of mood and cognitive and behavioral symptoms that show significant clinical diversity. The field of neuropsychiatric research has made progress but doctors continue to depend on clinical observations and patient self-reports for diagnosis because there are no established biological or computational biomarkers for automated decision support. The process of psychiatric diagnosis through subjective evaluation results in delayed diagnosis times which span from one year to several years before patients receive their official diagnosis thus causing their treatment outcomes to worsen and their life quality to decline[1-3]. Researchers use machine learning (ML) and deep learning (DL) techniques to build automated systems which perform psychiatric disorder assessments and predict symptom patterns. Researchers have used ML to analyze neuroimaging data from structural magnetic resonance imaging (sMRI) and functional MRI (fMRI) and electroencephalography (EEG) during the last ten years to develop schizophrenia identification tools. The diagnostic accuracy of these multimodal approaches reaches between 74% and 99% through the use of discriminative features which include cortical thickness and gray-matter volume and functional connectivity patterns and cross-modal data integration[4-8]. Such research shows that artificial intelligence (AI)-based analysis methods identify concealed brain patterns which distinguish patients from normal people thus enabling the creation of objective psychiatric diagnosis methods[1-3,5]. Most ML and DL models show promising accuracy but they function as unexplainable black-box systems which generate highly accurate results without showing their decision-making processes. Medical staff develop distrust toward AI tools because they cannot understand the operational mechanisms of these systems. The classification process in psychiatric disorders requires equal importance to both model understanding and the actual classification system. The model needs to reveal its decision limits and evidence basis and confidence level to help clinicians make responsible and accountable decisions[9-12]. The fundamental knowledge gap needs explainable AI (XAI) to operate as a multidisciplinary field which reveals model operations while making them understandable to humans. The Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) techniques provide feature influence explanations through basic visualizations which display the main factors that determine results. XAI methods in schizophrenia research have shown that the caudate nuclei and frontal operculum cortex and temporal fusiform cortex play essential roles in diagnosis while demonstrating ability to merge behavioral data with neuroimaging information[5,13-16]. The research demonstrates XAI serves two essential purposes by making scientific processes more transparent and by improving doctors' ability to understand AI-based medical systems. The current state shows major obstacles which prevent successful algorithm implementation in medical practice. The raw XAI outputs which include ranked feature lists and attribution heatmaps do not provide clinical value and cannot be easily used by psychiatrists during their decision-making process. The complex nature of schizophrenia which combines various neurobiological factors with multiple symptom presentations under one diagnosis makes it difficult to apply models between different groups and environments thus needing multiple tests for validation[1-3,5,9-16]. The fast development of AI technology and explainable modeling systems has become vital for schizophrenia diagnosis and clinical treatment. The research performs a systematic review to achieve three primary goals which involve: (1) Assessing ML and XAI systems for neuroimaging-based medical diagnosis; (2) Analyzing recent models that use behavioral and psychiatric data; and (3) Identifying critical methodological and ethical and translational barriers that must be addressed for medical practice adoption. The review aims to establish the development route from AI algorithm advancement to digital psychiatry systems which provide both trustworthy and clinically useful results and interpretable outcomes. The review investigates schizophrenia through an analysis of current AI system operations. The review extends past traditional epidemiological and etiological and clinical frameworks to demonstrate how modern computational tools using ML and multimodal neuroimaging analysis and explainable AI systems break through traditional diagnostic challenges to establish new methods for personalized psychiatric care. The review combines different perspectives to confirm its recently discovered research findings.

GENERAL DEFINITION

A general definition presents an extensive explanation which describes the core elements of a particular concept. The definition establishes basic characteristics of the concept through its explanation. The definition lacks precision because it provides a basic understanding of the concept through its main elements. The definition serves as an entry point to understand the concept better while providing a basis to study it in greater depth. Schizophrenia exists as a complicated mental health condition which produces recurring cycles of worsening symptoms and improvement[17]. The condition shows positive symptoms through delusions and hallucinations and disorganized speech and behavior together with negative symptoms that include social withdrawal and anhedonia and lack of motivation. The condition produces additional symptoms which include attention problems and memory issues and executive function deficits that indicate broad cognitive and perceptual and behavioral disturbances. Schizophrenia creates a severe impact on both people who have it and their families and the entire community[18,19]. The traditional clinical descriptions of schizophrenia together with its epidemiological patterns and etiological mechanisms and diverse clinical expressions serve as the essential base for understanding this disorder. The same features which define schizophrenia lead to ongoing diagnostic and therapeutic obstacles for clinicians. AI-driven analytical methods now use epidemiological data and neurobiological indicators and symptom patterns to create predictive models which enhance both early diagnosis and risk assessment and personalized treatment strategies. The conventional knowledge base serves as both historical background and structural base which enables AI systems to generate more precise data-based understanding of schizophrenia.

EPIDEMIOLOGY

Research studies in the literature show that schizophrenia exists in 0.75% of worldwide population according to epidemiological data[20]. The condition affects 21 million people worldwide according to estimates which place it within the schizophrenia spectrum disorders category[21,22]. Research studies have demonstrated that schizophrenia prevalence rates differ between different geographic areas with rates between 8 cases and 43 cases per 100000 population[22]. The worldwide number of schizophrenia patients amounts to 21 million people according to recent meta-analytic research which predicts that population aging will lead to increased numbers of affected individuals. Most patients live in developing nations which continue to experience major gaps in their treatment options[23]. The exact measurement of schizophrenia prevalence continues to be difficult because doctors struggle to make accurate diagnoses and because the condition shares symptoms with other mental health conditions and research methods vary between studies. Given these challenges, schizophrenia and related psychotic disorders are often grouped together in prevalence research. The World Health Organization (WHO) states that schizophrenia and related psychotic disorders occur in 0.25% to 0.64% of the United States population according to research from household surveys and medical records and structured diagnostic interviews[24-27]. Research studies have shown that this condition affects 0.33% to 0.75% of people worldwide[22,28-30]. In the literature, studies using AI models have been conducted to determine the prevalence of schizophrenia[8,29]. In one conducted study, the prevalence in the general population was reported as 1.6%. This rate was calculated to be somewhat higher than what is found in the literature. This situation has been attributed to various factors such as AI systems estimating lifetime prevalence rather than point prevalence and potentially using a more comprehensive disorder classification methodology[8].

The disorder typically starts affecting people between ages 15 and 25 before it leads to early death and suicide attempts become more common[30]. The WHO reports that schizophrenia first appears in people between their late teenage years and their early thirty's. The first signs of schizophrenia appear during late adolescence through early twenties in males but females show symptoms between their early twenties and early thirties[29,31,32]. Gender-related differences in schizophrenia have been recognized since the earliest descriptions of the disorder. Women develop their symptoms at an older age and their symptoms become more severe after they reach menopause. Although the incidence is higher in males, overall prevalence does not differ significantly between sexes. These disparities have been attributed to genetic and neurodevelopmental factors, as well as the potential neuroprotective effects of estrogen in women[33]. In the results obtained using AI systems in this field, it has also been shown that the disorder is more common among individuals who are single or separated from their partners, and that patients have difficulty working[8].

Research findings indicate that schizophrenia affects unmarried people at higher rates and people with schizophrenia experience difficulties when trying to build or sustain marriages[34-36]. Research findings show that people with schizophrenia tend to marry less often and face difficulties in maintaining their romantic partnerships. The condition leads to negative impacts on work performance because patients face challenges when looking for work and keeping their existing jobs[37-39]. The societal and economic burden of schizophrenia is substantial[20]. Employment rates among individuals with schizophrenia are remarkably low; studies in the United States and Europe report that only about 20% of patients hold regular paid employment[20,23,40]. The socioeconomic burden of chronic disorders continues to rise because patients experience high unemployment rates and their caregivers also face increased joblessness[41]. The United States spends 155.7 billion USD annually on schizophrenia treatment which includes 38% from unemployment costs and 34% from caregiver productivity losses and 24% from direct healthcare expenses[41].

Schizophrenia creates economic problems because it reduces worker productivity during their most productive years while also causing family members to lose productivity and forcing patients to pay for their own treatment and rehabilitation costs[23]. The total economic burden increases because emergency visits happen frequently and patients need hospital care and outpatient follow-ups and pharmacological treatment and extended rehabilitation periods[42,43]. Consequently, schizophrenia remains one of the most costly psychiatric disorders for both individuals and society[20,23].

ETIOLOGY

The exact causes of schizophrenia have not been fully determined because the condition develops from multiple genetic and neurobiological and environmental elements. Research shows that psychosis emerges through the combination of genetic factors with environmental elements which affect neurotransmitters and inflammation and birth complications and advanced paternal age and traumatic events and poverty and social isolation and migration and substance abuse[43-48]. In recent years, a considerable number of studies have been conducted using AI systems to elucidate the etiology of the disorder. In these studies, AI models have been developed based on various biomarkers such as gene signals, brain imaging, and inflammatory markers[49-53]. Data from the 2014 Psychiatric Genomics Consortium Schizophrenia Working Group identified 128 independent genetic variants across 108 chromosomal loci associated with schizophrenia[43]. Family and twin research studies demonstrate that genetic elements dominate development while environmental elements have little influence[44], The research findings do not support the expected pattern that was predicted. The research by Torrey and Yolken[44] demonstrated that monozygotic twins developed schizophrenia together at a 28% rate but dizygotic twins shared the condition at 6% which suggests environmental elements contribute to schizophrenia development[43]. The disorder results from multiple genetic variants which collectively contribute to susceptibility rather than being caused by a single gene or allele[45,46]. Polygenic risk scores constructed from schizophrenia-related alleles explain only about 7.7% of the variance in disorder liability[45]. In a study using AI modeling in this field, it was concluded that a selected group of candidate genes has great potential for uncovering the etiology[53].

Research shows prenatal and perinatal complications serve as established environmental risk factors yet scientists have not identified their exact mechanisms of action. Research has established that low birth weight and springtime births and maternal substance or alcohol exposure during pregnancy and paternal age above 40 and childhood experiences of neglect and abuse and peer bullying are major risk factors[18]. Research findings show that obstetric complications together with fetal development abnormalities increase schizophrenia risk at a moderate level[18,47,48]. Research shows that people born during late winter and early spring face higher health risks because of seasonal factors[54,55].

Schizophrenia develops through neurochemical disruptions which affect both dopaminergic and glutamatergic signaling pathways. Studies indicate that too much activity in the mesolimbic dopaminergic system results in the formation of positive psychotic symptoms[56]. Abnormalities in GABAergic and glutamatergic neurotransmission and an imbalance between these systems are increasingly recognized as contributors to both positive and negative symptom domains[57]. In addition, serotonergic, muscarinic, and glycinergic systems have been explored in recent research as potential modulators of the disorder[58,59]. Research now focuses on neuroinflammatory processes because scientists believe pro-inflammatory cytokines and oxidative stress markers play a role in causing disorder symptoms and disorder advancement[18].

Neuroimaging data shows that neuroanatomical abnormalities exist throughout the brain because of gray matter reduction and global brain volume loss and ventricular enlargement[60]. The lateral ventricles show enlargement while the cortex becomes thinner and the brain hemispheres lose their normal symmetry according to multiple studies[60-63]. Pronounced volumetric reductions are observed in the hippocampus, frontal, and temporal regions[64,65]. The presence of structural changes becomes evident through two lines of evidence which show decreased gray matter density and disrupted neural connections between neurons[66]. Functional imaging research supports these results by showing prefrontal cortical activity decreases which scientists refer to as hypofrontality[67-71]. In studies conducted with ML models, it has been shown that AI systems developed using different brain imaging techniques may play an important role in the diagnosis of schizophrenia[8]. In a meta-analysis, when various intracranial imaging and signal/signal-processing methods were used together with AI techniques to diagnose the disorder, a sensitivity and specificity of 80% were identified[69].

CLINICAL SYMPTOMS, FINDINGS, AND DIAGNOSTIC CRITERIA

Schizophrenia exists as a persistent mental health condition which produces multiple clinical symptoms including psychotic episodes and reality distortion and social isolation and motivational loss and reduced thinking capacity and neurocognitive deficits in attention and memory and executive abilities. The condition presents with positive symptoms which include hallucinations and delusions and disorganized speech and behavioral disturbances that result from abnormal function amplification and negative symptoms which cause decreased emotional and behavioral abilities through flat affect and reduced speech and motivation and social isolation and general functional decline[18].

Research findings indicate that psychosis patients exhibited unusual conduct during their pre-illness period through their academic and social interaction difficulties[70-72]. The initial symptoms typically emerge during late adolescence or early adulthood[23]. Research data shows that people with schizophrenia have homicidal behavior and suicide attempt rates at 0.48% to 0.75% which exceeds general population statistics[73-76].

In clinical practice, the diagnosis of schizophrenia is established based on the DSM-5 criteria (Figure 1), The diagnosis requires three elements which include specific symptoms and time duration and functional impairment.

Figure 1 Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition schizophrenia diagnostic criteria.

Schizophrenia has been linked to relapse and remission and chronic disorder patterns throughout its historical development[74]. Research studies published in the literature showed that 16.4% of patients recovered from their illness without any sex-related differences and found that recovery rates were higher in developing nations[75]. Research conducted in a meta-analysis study showed that 24.2% of patients reached recovery status while 35.5% achieved good or improved results and 59.7% received moderate outcomes[74]. The variability in these results is thought to be influenced by methodological differences across studies, variations in the stage and severity of illness among participants, and geographical differences in where the studies were conducted[74]. With the increase in AI studies in the field and the achievement of highly specific results, it has been shown that ML models may be effective in diagnosing schizophrenia, identifying clinical symptoms, and developing personalized treatment models in the coming years[2,76].

TREATMENT

The main goal of schizophrenia treatment involves symptom reduction to help patients regain their functional abilities and reintegrate into their community. Therapy success depends on physicians building strong collaborative bonds with patients through respectful communication and trust based relationships and open dialogue. The main treatment approach involves pharmacological interventions which use antipsychotic medications as their primary therapeutic option[73].

Typical antipsychotic drugs

The main action of typical antipsychotics occurs through D₂ receptor blockade in the mesolimbic pathway which reduces excessive dopamine activity that causes positive psychotic symptoms[77] (Table 1). The first-generation agent chlorpromazine and haloperidol and fluphenazine effectively treat hallucinations and delusions yet they cause extrapyramidal side effects and tardive dyskinesia in many patients.

Table 1 Commonly used antipsychotic medications.

Typical antipsychotics (first-generation)	Atypical antipsychotics (second-generation)
Phenothiazines: Chlorpromazine, fluphenazine, mesoridazine, perphenazine, thioridazine, trifluoperazine	Atypical agents: Aripiprazole, clozapine, olanzapine, quetiapine, risperidone, ziprasidone
Butyrophenones: Haloperidol
Others: Loxapine, molindone, thiothixene

Atypical antipsychotic drugs

The second-generation antipsychotics known as atypical agents form a diverse group which shows different receptor binding patterns and side effects and chemical properties compared to first-generation antipsychotics. These medications exhibit a higher affinity for serotonin (5-HT₂A) receptors relative to dopamine D₂ receptors, which contributes to their lower incidence of motor side effects and their efficacy in addressing both positive and negative symptoms[77] (Table 1). The treatment of schizophrenia includes aripiprazole and clozapine and olanzapine and quetiapine and risperidone and ziprasidone as standard medications.

The therapeutic response to antipsychotic medication may begin within days but often requires four to six weeks for full effect. Research studies through meta-analyses show that patients achieve their best treatment results during the initial seven to fourteen days of medication[77]. The treatment approach for patients who do not respond to standard antipsychotic doses and durations includes either increasing the medication dose or selecting a different antipsychotic medication[78]. The evaluation of medication adherence represents a vital process because depot formulations enable patients to maintain their treatment schedule when doctors identify non-adherence problems[79]. The recommended use of depot antipsychotics should be reserved for patients who fail to follow their medication schedule and for doctors who want to verify patient compliance with their medication regimen. The literature is increasingly highlighting that ML-based personalized treatment models may be used for many psychiatric disorders, including schizophrenia[80-82].

Electroconvulsive therapy

Electroconvulsive therapy (ECT) serves as an additional treatment for schizophrenia patients who do not respond to standard treatments or experience severe psychotic episodes. ECT produces its best results for patients who have recently developed their illness and show both sudden symptoms and strong emotional changes. Research studies show that ECT treatment leads to major short-term improvements for patients who experience their first schizophrenia episode[83,84].

COURSE, OUTCOME, AND PROGNOSIS

The prognosis of schizophrenia shows different outcomes for different people because of multiple biological and psychological and social elements that affect the condition. The outcome of disorder depends on four essential factors which include premorbid personality traits and family structure and social support and overall functional status. The prognosis for patients becomes worse when they show schizoid or schizotypal personality features and poor premorbid functioning and limited social support. The table shows that better long-term results occur when patients receive early and continuous treatment and when they have strong family connections and social networks and normal cognitive abilities (Table 2)[85].

Table 2 Prognostic indicators in schizophrenia.

Favorable prognostic indicators	Unfavorable prognostic indicators
Female sex	Male sex
Being married	Being single, divorced, or living alone
Acute onset	Insidious onset
Later age of onset	Early age of onset
Family history of mood disorder	Family history of schizophrenia
Good premorbid social adjustment	Premorbid schizoid or schizotypal personality
Living in a rural area	Living in an urban environment
Higher intelligence level	Lower intelligence level
Few and brief exacerbations	Frequent or prolonged hospitalizations
Early initiation and regular adherence to treatment	Irregular or absent treatment adherence
Absence of comorbid psychiatric disorders	Presence of comorbid psychiatric conditions
Strong family collaboration with the treatment team	Dysfunctional family environment or high expressed emotion

The illness of schizophrenia progresses through a long-term pattern of recurring episodes which alternate between times of improvement and times of worsening symptoms. The treatment and support of patients lead to satisfactory functional recovery for one-third of patients but two-thirds experience partial improvement with ongoing symptoms and the last third develops enduring disability and social integration issues. Research indicates that patients who get early diagnosis and continuous medication and psychological support and build strong doctor-patient relationships will achieve improved long-term outcomes with reduced relapse rates. The prognosis of schizophrenia worsens when patients fail to follow their treatment plan and when their condition starts before age 20 and when they are male and when their family environment shows high levels of stress and expressed emotion.

AI-ASSISTED APPROACHES TO SCHIZOPHRENIA DIAGNOSIS

The implementation of AI technologies in psychiatry during the last few years brought about a major shift in schizophrenia biomarker detection and clinical decision support systems. AI processes complex neurobiological and cognitive and behavioral data to achieve better diagnostic results and predict treatment outcomes and start interventions at an early stage. AI-driven models analyze neuroimaging data together with EEG recordings and speech patterns and social behavior measurements to create a complete understanding of schizophrenia's neurophysiological and phenotypic diversity. The AI-assisted methods enable biomarker-based diagnostic system development and improve clinical decision-making through XAI principles which provide explainable AI.

The rise of digital psychiatry

The diagnosis of complex psychiatric disorders such as schizophrenia has long relied on subjective clinical observations based on DSM-5 criteria. The practice of relying on clinical judgment leads to different treatment approaches between healthcare providers while the lack of biological markers makes it difficult to diagnose patients correctly and create proper treatment plans[86,87]. The diverse symptoms of schizophrenia along with its unknown brain mechanisms require scientists to create new diagnostic methods which can measure symptoms objectively. The new digital psychiatry paradigm uses AI to analyze complex clinical and neuroimaging and neurophysiological data for early disorder detection and individualized treatment development[88,89]. AI systems process EEG and MRI data together with genomic information and behavioral metrics and electronic health records to create biomarker-based data-driven decision support systems for schizophrenia and psychiatric disorders[90,91]. The fast progress of digital psychiatry models has not solved the essential problem of making their operations transparent and easy to understand. Establishing trust in clinical decision-making and ensuring the ethical adoption of AI systems require that model reasoning processes be explainable. The development of XAI methodologies works to enhance model output understanding which leads to better clinical decision-making confidence[92,93]. Specifically, SHAP quantifies the contribution of individual input features to model decisions; Gradient-weighted Class Activation Mapping visualizes salient regions in DL networks; and LIME generates localized explanations for individual predictions. The combination of these methods enables AI models to show their causal reasoning for schizophrenia detection and other neuropsychiatric classifications which enhances clinicians' ability to interpret results[94-96]. Digital psychiatry brought a fundamental change to medical practice because doctors now use data analysis to make diagnoses instead of depending on their personal opinions. The combination of AI with explainable modeling systems produces better diagnostic results while creating an AI-assisted mental health system which operates ethically and provides individualized care through human-AI teamwork[97-99].

Neuroimaging and AI based approaches in schizophrenia

The research of schizophrenia depends on neuroimaging techniques which use MRI and fMRI and diffusion tensor imaging (DTI) as their core methods. Research using structural MRI has shown that schizophrenia patients develop smaller brain volumes in their hippocampus and prefrontal cortex starting from the beginning of their illness. The disorder produces significant brain structure abnormalities which scientists have identified through these studies[100,101]. Functional MRI investigations have further shown that disruptions in large-scale neural networks particularly the default mode and salience networks are closely linked to cognitive dysfunctions and impaired integration of neural information[100,102]. Complementing the research findings show that DTI analyses detected white matter integrity breakdowns in brain connections between frontal and temporal areas which indicates neural communication problems and decreased brain information processing speed[100,103].

AI and EEG based approaches in schizophrenia

The combination of AI with EEG technology now serves as a revolutionary method for schizophrenia diagnosis and assessment and ongoing monitoring. The system enables researchers to detect complex neurophysiological patterns in EEG data through an automated non-invasive method which replaces traditional behavioral observation and subjective clinical evaluation. AI technology helps researchers analyze complex EEG data to study schizophrenia brain abnormalities which leads to better detection methods and individualized treatment plans. EEG recordings from schizophrenia patients show continuous disruptions across different frequency bands which extend throughout their entire recording duration. Research has shown that theta (4-7 Hz) and alpha (8-11 Hz) and beta (12-30 Hz) and gamma (> 30 Hz) power bands show abnormal increases which mainly affect the temporal and posterior cingulate regions[104,105]. Research shows that middle temporal gyrus theta activity at elevated levels creates a negative relationship with verbal memory abilities which proves EEG signals directly affect cognitive performance[105,106]. Research indicates that brain region synchronization problems result in phase discontinuity which represents a basic neurophysiological brain function defect. The brain shows two main patterns of disruption through its neural communication system because alpha-phase discontinuity increases and theta-phase continuity decreases. The alpha peak frequency in temporal and cingulate regions shows a substantial decrease which scientists identify as a distinct marker of the disorder[105,106]. The brain responses known as event related potentials (ERPs) show two types of abnormalities in schizophrenia: P50 gating and mismatch negativity which link to social and cognitive impairments[106]. Research has demonstrated that AI achieves successful EEG-based schizophrenia classification through its application of ML and DL frameworks. The classical algorithms support vector machines (SVM) and decision trees and random forests and logistic regression show effective interpretable results in small datasets where decision trees reach 89.12% specificity and random forest models achieve 91% specificity[107,108]. DL systems use their operation to find both time-based and space-based patterns which exist in EEG data. The system employs convolutional neural networks (CNNs) to detect particular patterns between channels and long short-term memory (LSTM) networks to detect time-dependent patterns. The hybrid CNN-LSTM models achieve 99.25% classification accuracy according to[109,110]. The ten-layer CNN models which process multiple EEG channels achieve 99.18% accuracy but GRU networks that focus on alpha rhythms reach 88.88%[110]. The method needs to undergo multiple site testing for clinical use because it requires both strong validation results and established measurement protocols. The study used 81 participants from various educational institutions and 80 schizophrenia patients to obtain reliable ERP and quantitative EEG results which showed connections to cognitive performance and functional assessment results[111]. EEG-based AI systems require sophisticated preprocessing and feature extraction methods to generate reliable analytical results. The signals undergo delta theta alpha beta and gamma band decomposition after band-pass filtering and noise reduction. The mathematical feature extraction methods of fast Fourier transform, approximate entropy, log-energy entropy, Shannon entropy, and kurtosis are commonly used, with entropy measures quantifying signal complexity and irregularity, and kurtosis reflecting the shape of the amplitude distribution[112]. The model achieves better performance through normalization techniques which use z-score and L2 normalization to distribute data values evenly[109]. The combination of Bayesian optimization with recursive feature elimination helps prevent overfitting in high-dimensional EEG data which results in better model performance[109]. AI-assisted EEG analyses demonstrate the ability to identify pre-conditions which appear before the development of the condition. People in this stage show cognitive problems which they cannot measure but these symptoms do not reach the level needed for psychotic disorder diagnosis. Research conducted with the Bonn Scale for the Assessment of Basic Symptoms demonstrates that psychosis detection at an early stage prevents its development in people who experience it[113,114]. AI models gain improved disorder onset prediction during the prodromal stage through the combination of EEG biomarkers with natural language cues from clinical notes that include symptoms like “hearing voices” and “social withdrawal”[115]. Healthcare professionals require model interpretability as an essential factor to accept these models for clinical use. The DL system operates as a black box which makes it difficult for clinicians to understand diagnostic results so they need XAI methods including SHAP and LIME for explanation[15]. The methods enable researchers to visualize EEG features together with brain regions and connectivity patterns which drive classification results for better model understanding. XAI methods in research studies identified age and gender and spectral EEG patterns as the essential diagnostic factors for schizophrenia diagnosis. The SHAP analysis showed that schizophrenia classification depends on three brain regions which include the bilateral pallidum and inferior frontal gyrus (pars triangularis) and temporal regions[1,5,15]. Schizophrenia neurobiological origins exist beyond EEG because EEG combined with fMRI and DTI and MR spectroscopy and MEG allows researchers to study structural and functional and electrophysiological aspects at the same time[1,5,15]. AI-based multimodal frameworks use gray matter volume loss and functional connectivity reductions and N-acetylaspartate depletion and EEG rhythmic disturbances to create personalized biomarker profiles which improve treatment prediction accuracy[116,117]. The new field of precision psychiatry requires these intricate evaluation methods for its operation. AI systems use EEG biomarkers together with genetic information and molecular data and environmental factors to develop individualized neurobiological subtypes which direct personalized treatment approaches[118,119]. Nonetheless, many existing studies are limited by small sample sizes, leading to overfitting and reduced external validity[118,120]. The use of pharmacological treatments in treatment allows disorders to produce specific EEG patterns which remain undetectable to algorithms because Western researchers collect most of the available data[108,121,122]. The application of transfer learning and domain adaptation methods has proven successful for improving generalization performance when working with restricted EEG datasets[123,124]. The combination of CNN, LSTM and transformer architectures in hybrid models produces better results in terms of accuracy and interpretability and attention mechanisms help detect important EEG segments for diagnosis[125]. AI-assisted EEG applications provide schizophrenia researchers with objective neurophysiological indicators which help them detect the condition at its initial stages and throughout its progression. The CNN-LSTM-based models reached 99% accuracy which demonstrates the potential of EEG-based methods for diagnosis through their ability to detect theta, alpha and gamma band irregularities and phase disruptions and ERP changes and theta-band coherence variations[106,109,110,126]. The clinical validity and reliability of these technologies need to be proven through extensive multicenter research that includes diverse participants and follows uniform EEG testing methods and implements clinical decision support systems with XAI. AI-driven EEG analysis methods have the ability to create an early diagnostic system for schizophrenia which provides personalized and objective psychiatric assessments.

Explainability and clinical integration

XAI represents a new method which has established itself as an essential tool for schizophrenia diagnosis and treatment in medical settings. The psychiatric disorder schizophrenia depends on clinical observation and patient self-reports for diagnosis because its neurobiological mechanisms are complex. AI technologies show promise to improve diagnostic precision and increase early detection chances through biomarker-based decision support systems[1,127]. XAI methodologies enhance model decision interpretation which results in higher clinical trust. The techniques of LIME and SHAP and ELI5 help doctors understand DL model decisions by showing which input features affect the output results[15]. The combination of fMRI with AI technology enables researchers to detect prefrontal cortex and hippocampus abnormalities which helps them understand how default mode network connectivity problems lead to cognitive impairments[12]. The BiLSTM-Attention DL models which use EEG data enable researchers to visualize temporal neurophysiological features in an interpretable way which enhances medical professionals' ability to understand model results[15]. Natural language processing-based systems help detect symptoms through their analysis of speech content and fluency patterns. The UCL Institute of Neurology conducts research which demonstrates that language model analysis helps scientists detect semantic in-coherence and reduced verbal predictability in schizophrenia patients relative to healthy participants[128]. Research studies have shown that AI-based diagnostic systems reach 99.25% classification accuracy for schizophrenia detection according to recent systematic reviews[127,129]. The new method produces better results than conventional clinical tests because it achieves results that traditional methods cannot match since they produce 55%-76% incorrect diagnoses. The diagnostic and treatment planning capabilities of SVM and Random Forests and Vision Transformers have shown significant promise in medical applications[12,127]. Multiple major obstacles exist during XAI implementation in clinical practice which affect operational use and generate ethical and legal problems. The current models struggle with generalizability because they depend on limited datasets which fail to include diverse demographic populations. The training of models using Western population data leads to poor performance in different cultural and ethnic groups which creates algorithmic bias that produces clinical mistakes[12,115,130,131]. The internal operations of DL models remain impossible to understand which prevents medical staff from tracking their decision-making process. XAI techniques help solve this problem but their generated explanations stay complex and difficult to use in clinical decision support systems. The implementation of AI systems faces opposition from mental health professionals because they worry these systems will replace human clinical judgment which leads to resistance against technology adoption[12,115,130,131]. Mental health treatment requires two fundamental ethical principles which include informed consent and data privacy and confidentiality. Legal accountability in cases of misdiagnosis or detection of high-risk behaviors (e.g., suicidal ideation) known as the Tarasoff duty remains unresolved[115]. The absence of full datasets which show balanced information through ethical representation makes it difficult to compare explainable models and establish standards. The existing datasets including DAIC-WOZ and CLPsych and AVEC face multiple restrictions which affect their compliance with DSM-5 standards and their interpretability and their ability to represent diverse demographics and their stability across time[13]. The European Union Artificial Intelligence Act (2024) identifies schizophrenia diagnosis AI systems as high-risk so organizations need to establish complete transparency and accountability systems. The Canadian government has established the Artificial Intelligence and Data Act to create similar regulatory systems. The United Kingdom operates the Digital Technology Assessment Criteria through the National Health Service to enforce requirements for clinical safety and data protection and system accessibility[132,133]. The future of clinical integration depends on developing explanations which adjust to particular situations and user requirements and support AI-clinician dialog systems and social-cognitive abilities and uphold professional responsibility[134]. XAI requires explanations which should align with clinical environments and patient requirements and user skill levels to succeed in clinical settings. The international community has established accountability and transparency and fairness and person-centeredness and data protection and security as fundamental ethical principles for AI mental health applications[132,133]. The upcoming research will develop hybrid systems which unite biological information with behavioral data and neuroimaging results. AI models that explain their operations by uniting genetic polymorphisms with serum biomarkers and behavioral indicators and structural brain data will generate better and more useful predictions[135,136]. The diagnostic and treatment capabilities of schizophrenia will experience a revolutionary change through explainable AI because it enables better early detection and individualized care and risk assessment. The potential for AI in clinical practice will become achievable through better data quality and established ethical and legal frameworks and clinician knowledge of AI systems and standardized validation procedures. XAI technology integration into schizophrenia treatment brings more than new technology because it creates a complete system which unites human-centered ethical principles with dependable medical progress.

CONCLUSION

In the AI era, understanding schizophrenia requires integrating established clinical knowledge with emerging computational tools that increasingly redefine how the disorder is detected, characterized, and managed. The current diagnostic methods depend on doctor observations and patient feedback which hinders both early disorder detection and personalized medical treatment development. The development of psychiatric research now depends on AI methods which serve as its primary driving force. AI performs objective schizophrenia analysis through multiple biological indicators which combine EEG and MRI and fMRI and DTI and genomic and behavioral and linguistic data to enhance diagnostic accuracy and develop reliable clinical prediction systems. The development of XAI methodologies has accelerated this transformation by enabling medical professionals to understand model decisions at a clinical level which leads to the development of human-centered decision-support systems that maintain trustworthiness. Research data indicates AI systems using EEG data can precisely detect neurophysiological biomarkers and neuroimaging methods generate brain structure and function models of abnormalities and AI systems evaluate social and speech data to assess schizophrenia-related language and behavioral patterns. Multimodal AI frameworks merge various data modalities into a single analytical domain which enables researchers to develop customized disorder profiles. The current foundation enables AI predictive technology to develop precise psychiatry through personalized treatments which analyze neurobiological and behavioral patterns of each patient. Nevertheless, current literature reveals several important limitations. AI models have been created through research using restricted data sets which mostly stem from Western populations thus restricting their ability to work effectively in diverse cultural settings. The DL models operate as black boxes because they prevent medical staff from understanding their decision-making processes. XAI techniques solve part of this problem but researchers need to develop methods which convert algorithmic explanations into medical terms and create simple clinical interfaces for users. The ethical framework lacks solutions for data privacy protection and informed consent management and bias control and legal responsibility enforcement at a standardized global level. The research needs to progress through three fundamental directions for future studies. The development of extensive datasets which span multiple centers and include diverse cultural groups will improve model performance and clinical trustworthiness. The analysis will track disorder progression through a single analytical system which combines EEG data with fMRI results and linguistic and behavioral indicators. The medical field needs standardized ethical and explainable AI frameworks as its most critical requirement to achieve responsible AI system deployment in the future. The European Union Artificial Intelligence Act along with other international regulations require clinical AI systems to follow principles which include transparency and accountability and fairness. AI technology implementation in schizophrenia research brings new tools but also transforms the entire method of psychiatric scientific research. AI models that explain their decisions while following ethical standards and working as complete systems will enable clinicians to receive strong assistance for both disorder detection at the beginning and treatment outcome forecasting and symptom tracking and rehabilitation planning. The success of this transformation depends on accurate algorithms and human-centered design approaches and ethical practices and teamwork between different fields of expertise. The review shows that AI systems which provide explainable functions and follow ethical guidelines will enhance diagnostic precision and develop personalized treatment approaches which result in superior long-term outcomes for schizophrenia research and clinical practice.