Neuroradiological features across major psychiatric disorders and behavioral variant of Alzheimer’s disease: a concise multimodal synthesis

Abstract

Psychiatric disorders are complex, disabling conditions that continue to rely on subjective diagnostic criteria due to the absence of objective biological markers. Neuroradiology has become a critical discipline for examining the structural, functional, and biochemical underpinnings of these disorders through advanced brain imaging. This review synthesizes findings from five major psychiatric conditions including major depressive disorder, schizophrenia, autism spectrum disorder, obsessive-compulsive disorder, and generalized anxiety disorder, and briefly discusses the behavioral variant of Alzheimer’s disease, a variant with neuropsychological overlay, across multiple imaging modalities, including structural magnetic resonance imaging (MRI), diffusion tensor imaging, functional MRI, magnetic resonance spectroscopy, functional near-infrared spectroscopy, and positron emission tomography. We present a comparative overview of cross-condition and modality-specific findings, highlighting converging disruptions in frontolimbic and temporoparietal circuits, alongside unique neurobiological features in each disorder. We also acknowledge key confounds such as medication effects, comorbidities, and methodological variability that limit direct transdiagnostic inference. We further discuss methodological limitations, emerging trends such as multimodal integration and machine learning, and future directions for translating imaging data into clinically meaningful biomarkers.

Introduction

Psychiatric disorders are among the leading causes of disability worldwide, affecting hundreds of millions of individuals and significantly burdening healthcare systems (Eaton et al., 2008). Despite their impact, diagnostic frameworks remain rooted in clinical interviews and behavioral observation, which, while essential to psychiatric practice, are inherently variable and sometimes imprecise (Tan et al., 2023). The absence of reliable biological markers complicates diagnosis, subtyping, prognosis, and treatment selection, leading to trial-and-error pharmacotherapy and inconsistent outcomes (Abi‐Dargham et al., 2023). Neuroimaging, by contrast, provides an objective window into the biological substrates of psychiatric illness, with the potential to refine diagnosis and guide individualized therapeutic strategies.

The modern neuroradiological study of psychiatric disorders can be traced to the 1970s, when computed tomography (CT) revealed ventricular enlargement in schizophrenia (Johnstone et al., 1976). Since then, the field has expanded to include magnetic resonance imaging (MRI) and advanced modalities such as diffusion tensor imaging (DTI), magnetic resonance spectroscopy (MRS), and positron emission tomography (PET). This integration of neuroimaging into psychiatric research has been referred to as psychoradiology (Lui et al., 2016); however, in this review we use the term neuroradiology of psychiatric disorders to emphasize both its clinical grounding and methodological scope.

Despite notable progress, neuroimaging findings across psychiatric research remain heterogeneous, influenced by sample variation, methodological differences, and limited replication. Prior reviews often examine single disorders or modalities, reducing opportunities for cross-diagnostic comparison and limiting our understanding of shared neurobiological mechanisms. An integrative framework that examines structural, functional, and molecular signatures across disorders may help position psychiatric conditions along a broader neurobiological continuum spanning developmental, neurochemical, and degenerative pathways.

Within this framework, the behavioral variant of Alzheimer’s disease (bvAD) serves as a valuable neurodegenerative comparator. Although bvAD is driven by amyloid- and tau-mediated pathology, it frequently presents with apathy, disinhibition, emotional blunting, and impaired social cognition, features that substantially overlap with major psychiatric disorders (Seeley et al., 2009; Rascovsky et al., 2011; Lanctot and Fink, 2023). This convergence reflects dysfunction within common frontolimbic and salience-network circuits, demonstrating how distinct upstream mechanisms can yield similar network-level disruptions (Filippi and Agosta, 2011; Pievani et al., 2014; Taylor et al., 2023). As such, bvAD functions as a mechanistic reference point, clarifying differences in etiology while illuminating shared patterns of circuit vulnerability across psychiatric illness.

This review synthesizes neuroradiological findings across major depressive disorder (MDD), schizophrenia, autism spectrum disorder (ASD), obsessive compulsive disorder (OCD), generalized anxiety disorder (GAD), and bvAD. These conditions were selected based on global prevalence, clinical impact, and the availability of multimodal neuroimaging evidence allowing meaningful cross-disorder comparison. A narrative, selective search strategy was employed across PubMed, Scopus, and Web of Science, prioritizing meta-analyses, collaborative datasets such as Enhancing Neuroimaging Genetics through Meta-Analysis (ENIGMA) and UK Biobank, and high-quality multimodal or transdiagnostic studies.

Neuroimaging modalities in psychiatric disorders

Structural magnetic resonance imaging

Structural MRI (sMRI) provides high-resolution anatomical detail, enabling quantification of cortical thickness, gray–white matter volume, surface area, and subcortical morphology (Etkin and Mathalon, 2024). This modality forms the foundation of psychiatric neuroimaging research due to its accessibility, reproducibility, and suitability for large-scale analysis. Multisite initiatives such as ENIGMA, the UK Biobank, and distributed computation frameworks like the Collaborative Informatics and Neuroimaging Suite Toolkit for Anonymous Computation (COINSTAC) have greatly expanded analytic power, supporting cross-diagnostic comparisons with unprecedented sample sizes (Sudlow et al., 2015; Gazula et al., 2020; Thompson et al., 2020).

sMRI is particularly valuable in longitudinal frameworks where it can quantify neuroanatomical trajectories, treatment-related change, and illness progression. Morphometric features, including ventricular enlargement, cortical thinning, and hippocampal atrophy, have been incorporated into machine-learning classification models with encouraging diagnostic accuracy (Zhang et al., 2025). Longitudinal research demonstrates that structural change is dynamic and clinically meaningful: in treatment-resistant depression, responders show increases in cortical thickness and hippocampal volume, whereas non-responders exhibit persistent or progressive reductions (Phillips et al., 2015). In bipolar disorder, progressive cortical thinning in frontal and temporal cortices co-occurs with relative preservation or thickening in sensory and visual areas; lithium treatment is consistently associated with gray-matter increases and structural stabilization (Lyoo et al., 2010; Abé et al., 2020). Additional evidence suggests that some morphologic abnormalities—particularly in major depressive episodes—may partially normalize with symptom remission, indicating reversible, state-dependent neural plasticity (Ahdidan et al., 2011).

Beyond regional morphology, network-level analysis of sMRI-derived structural covariance has revealed abnormalities in topological organization across psychiatric populations (Gonuguntla et al., 2022). To contextualize these findings quantitatively, large-scale analyses from the ENIGMA consortium have demonstrated moderate-to-strong spatial correlations among cortical-thickness alteration maps for schizophrenia, bipolar disorder, MDD, and OCD (r = 0.44–0.78), whereas correlations with attention-deficit/hyperactivity disorder (ADHD) and ASD are weak or absent (Opel et al., 2020). Across these disorders, standardized mean differences for regional cortical thinning generally fall within a small-to-moderate range (Cohen’s d ≈ –0.10 to –0.57), with the largest effects observed in frontal and temporal association cortices that support affective and cognitive control (Schmaal et al., 2017; Hibar et al., 2018; Van Erp et al., 2018; Boedhoe et al., 2020). Although these effect sizes are modest, their spatial consistency underscores shared structural network vulnerability across psychiatric disorders, forming a quantitative foundation for cross-modal comparisons. However, sMRI captures anatomy rather than neural dynamics or neurotransmitter function and is therefore most powerful when integrated with functional or molecular modalities (Etkin and Mathalon, 2024).

Synthesis of modality-specific findings

Across psychiatric illnesses, sMRI reveals both shared and disorder-specific patterns of macrostructural alteration. Convergent abnormalities include frontolimbic cortical thinning, hippocampal and amygdala volume reduction, and decreased integrity of prefrontal–temporal networks—changes that broadly align with impairments in emotional regulation, memory, salience processing, and executive function (Schmaal et al., 2017; Van Erp et al., 2018; Boedhoe et al., 2020). Disorder-specific profiles also emerge; schizophrenia and bipolar disorder show more widespread cortical thinning, particularly in prefrontal and temporal association cortices (Hibar et al., 2018; Van Erp et al., 2018), ASD demonstrates early-life cortical enlargement followed by atypical developmental trajectories (Ecker et al., 2015; Hazlett et al., 2017), and MDD often exhibits hippocampal and subgenual cingulate atrophy that may respond to treatment (Phillips et al., 2015; Schmaal et al., 2017) (Fig. 1). In bvAD, sMRI typically shows cortical thinning and gray matter loss within the orbitofrontal cortex, anterior cingulate gyrus, insula, and dorsolateral prefrontal cortex—regions essential for social cognition and emotional regulation. Hippocampal atrophy is present but often less pronounced in early disease stages, contrasting with the amnestic-dominant pattern of typical AD. This frontal-predominant atrophy pattern underlies the marked behavioral and dysexecutive features of bvAD and highlights how network-level degeneration can mimic the macrostructural signatures seen in primary psychiatric disorders (Whitwell et al., 2012; Ossenkoppele et al., 2015; Singleton et al., 2020). Collectively, sMRI findings across psychiatric and neurodegenerative spectra suggest convergent vulnerability of frontolimbic and salience-related networks, where distinct upstream mechanisms—neurodevelopmental, affective, or degenerative—manifest through overlapping structural end points.

Figure 1.

Volumetric abnormalities across six disorders. Heatmaps illustrate gray matter co-atrophy patterns, highlighting shared and disorder-specific structural alterations. MDD (1) and schizophrenia (5) show overlapping frontolimbic volume loss, ASD (2) and OCD (3) demonstrate more localized changes, GAD (4) involves limbic and paralimbic reductions, and bvAD (6) exhibits widespread frontoinsular, temporal, and parietal atrophy. Color intensity reflects the consistency of findings across studies, emphasizing broad transdiagnostic structural trends rather than quantitative regional measures.

Diffusion tensor imaging

Diffusion tensor imaging (DTI) enables in vivo characterization of white matter microstructure by measuring directional water diffusion. Quantitative parameters including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) reflect the degree and orientation of diffusion anisotropy arising from axonal membranes, myelin, neurofilaments, and microtubular organization (Podwalski et al., 2021). Because efficient neural communication depends on coherent tract architecture, DTI is particularly well suited for detecting subtle disruptions in major white matter pathways implicated in psychiatric illness (Kashefi and Winston, 2020).

Technological advances including multi-shell acquisition, high-field scanners, and improved tractography have enhanced resolution and expanded applications in connectomics and outcome prediction (Meoded et al., 2020). Although DTI has proven valuable for clinical stratification and symptom mapping, its broadest strength lies in quantifying macro-scale structural connectivity. However, the tensor model assumes Gaussian diffusion with a single dominant fiber orientation per voxel, an oversimplification in regions with crossing, branching, or “kissing” fibers. Furthermore, FA, MD, and eigenvalues are biologically nonspecific, reflecting multiple microstructural processes such as myelin loss, axonal degeneration, edema, and neuroinflammation. These limitations have driven the development of advanced diffusion models including diffusion spectrum imaging (DSI), neurite orientation dispersion and density imaging (NODDI), and constrained spherical deconvolution (CSD), which permit more accurate estimation of neurite complexity and orientation dispersion. Motion sensitivity and preprocessing variability remain challenges but can be mitigated through harmonized acquisition and multimodal integration (Assaf and Pasternak, 2008; Jones et al., 2013).

Synthesis of modality-specific findings

Across psychiatric populations, DTI consistently reveals reduced white matter integrity within fronto-limbic, salience, and default-mode networks, reflecting disruptions in long-range connectivity (Dong et al., 2017). MDD frequently shows decreased FA in the uncinate fasciculus and cingulum bundle (Liao et al., 2013; Bessette et al., 2014), while schizophrenia demonstrates more widespread dysconnectivity involving fronto-temporal tracts, superior/inferior longitudinal fasciculi, and callosal fibers (Wheeler and Voineskos, 2014; Kelly et al., 2018). ASD often exhibits early-life increases in diffusivity followed by developmental divergence, consistent with atypical maturation trajectories (Travers et al., 2012; Ameis and Catani, 2015). OCD and GAD show alterations in cortico-striatal and thalamo-prefrontal pathways, aligning with dysregulation of inhibitory control and threat-processing circuits (Tromp et al., 2012; Piras et al., 2013) (Fig. 2). In bvAD, diffusion imaging reveals microstructural disruption predominantly within frontotemporal white matter tracts, including the uncinate fasciculus and cingulum bundle, which connect orbitofrontal and limbic regions critical for emotional regulation and social cognition. These alterations parallel the tract-level dysconnectivity observed in primary psychiatric disorders but stem from neurodegenerative mechanisms such as tau-mediated axonal injury and myelin breakdown. Studies indicate that reduced FA in parahippocampal and frontal association fibers correlates with behavioral disinhibition and executive dysfunction, underscoring the shared vulnerability of frontolimbic circuitry across neuropsychiatric and neurodegenerative spectra (Whitwell et al., 2012; Brier et al., 2014; Ossenkoppele et al., 2015). Together, DTI findings suggest that despite divergent etiologies, both psychiatric and neurodegenerative conditions converge on a common architecture of white matter vulnerability involving frontotemporal and limbic pathways central to affective and cognitive control.

Figure 2.

DTI connectivity abnormalities across six disorders. Diffusion heatmaps display white matter tract disruptions within frontal, limbic, and subcortical pathways. MDD (1) and GAD (4) show reduced integrity in the uncinate fasciculus and cingulum, ASD (2) and schizophrenia (5) demonstrate frontoparietal and frontotemporal disconnection, OCD (3) involves cortico-striato-thalamo-cortical pathway disruption, and bvAD (6) shows degeneration in parahippocampal and frontoparietal tracts. Color intensity reflects agreement across diffusion studies.

Functional magnetic resonance imaging

Functional MRI (fMRI) detects fluctuations in blood-oxygen-level–dependent (BOLD) signals that serve as indirect markers of neural activity, either during externally driven tasks (task-based fMRI) or in the absence of explicit stimuli (resting-state fMRI). Both approaches are central to mapping large-scale functional networks, most notably the default mode, salience, and executive control networks, which underlie emotion regulation, attention, memory, and cognitive flexibility (Chen and Gauthier, 2021; Vachha et al., 2025). Resting-state fMRI has become particularly prominent due to its ease of acquisition and capacity to reveal intrinsic network dysconnectivity, even in patients unable to reliably perform tasks.

Recent advances in signal denoising, harmonized preprocessing pipelines, and access to large open datasets have improved robustness and reproducibility in network-level analyses (Di and Biswal, 2023). Graph-theoretical frameworks and dynamic connectivity analyses now support characterization of time-varying network states, addressing limitations of static connectivity models. fMRI is increasingly integrated into machine-learning workflows for diagnostic classification, subtype discovery, and treatment–response prediction (Miranda et al., 2021). Real-time fMRI neurofeedback (rt-fMRI-NF) represents an emerging translational extension, allowing participants to modulate neural activity based on live feedback—an approach with potential in disorders marked by dysregulated self-monitoring or affective control, though reproducibility and durability of therapeutic benefit remain under investigation (Pindi et al., 2021).

Despite its strengths, fMRI is constrained by indirect neurovascular coupling, motion sensitivity, variable hemodynamic response across individuals, and high cost. BOLD signals lack cellular specificity and may reflect vascular reactivity as much as synaptic activity. For these reasons, fMRI is most powerful when combined with structural and diffusion measures or when integrated with higher-order statistical modeling such as connectome-based predictive modeling, independent component decomposition, or multimodal data fusion approaches.

Synthesis of modality-specific findings

Across psychiatric disorders, fMRI reveals network-level dysregulation rather than isolated regional deficits. A consistent finding is default mode network (DMN) hyperconnectivity in depression, particularly in rumination-dominant subtypes, accompanied by weakened anticorrelation with executive control networks (Sheline et al., 2010; Kaiser et al., 2015). Dysfunction of the salience network is frequently observed in schizophrenia, bipolar disorder, and autism, reflecting impaired switching between internal and external cognitive states (Palaniyappan and Liddle, 2012; Uddin, 2015). Frontoparietal and cognitive-control network hypoconnectivity, common to depression, OCD, and anxiety disorders, aligns with deficits in inhibition, cognitive flexibility, and goal-directed behavior (Fitzgerald et al., 2017; McTeague et al., 2020), while corticostriatal abnormalities characterize compulsivity in OCD (Harrison et al., 2009; Beucke et al., 2013). Disorder-specific patterns refine this shared architecture: schizophrenia shows disrupted thalamo-cortical integration and aberrant temporal–parietal engagement (Woodward et al., 2012); autism often exhibits atypical developmental connectivity with long-range hypoconnectivity and local hyperconnectivity (Uddin et al., 2013); and anxiety disorders display heightened amygdala responsivity with insufficient prefrontal regulation (Etkin and Wager, 2007; Sylvester et al., 2012) (Fig. 3). In bvAD, resting-state fMRI demonstrates disrupted connectivity within the salience and frontoparietal control networks, along with altered interactions between the DMN and executive systems. This network-level profile parallels the dysconnectivity observed across major psychiatric disorders but reflects neurodegenerative mechanisms centered on tau-mediated hub disintegration rather than functional dysregulation alone. Such salience–DMN imbalance contributes to apathy, disinhibition, and impaired social cognition, reinforcing the concept of convergent circuit vulnerability across psychiatric and neurodegenerative conditions (Seeley et al., 2009; Ossenkoppele et al., 2015; Singleton et al., 2020; Lanctot and Fink, 2023). Collectively, fMRI findings across these domains support a transdiagnostic framework in which both psychiatric and neurodegenerative diseases arise from large-scale network instability affecting salience detection, executive control, and self-referential processing.

Figure 3.

Functional connectivity abnormalities across six disorders. Functional connectivity heatmaps depict network disruptions across disorders. MDD (1) and GAD (4) demonstrate reduced frontolimbic coupling, ASD (2) shows reduced long-range but increased local connectivity, OCD (3) displays cortico-striato-thalamo-cortical hyperconnectivity, schizophrenia (5) shows global hypoconnectivity, and bvAD (6) presents frontolimbic and default mode disconnection. Color intensity indicates relative consistency across the literature.

Magnetic resonance spectroscopy

MRS allows in vivo quantification of neurochemical metabolites by measuring the resonance frequencies of hydrogen nuclei. Core analytes include: N-acetylaspartate (NAA), commonly interpreted as a marker of neuronal integrity; creatine (Cr), reflecting cellular energy metabolism; choline-containing compounds (Cho), indexing membrane turnover and myelination dynamics; and glutamate (Glu) and gamma-aminobutyric acid (GABA), which constitute the brain’s principal excitatory–inhibitory transmitter pair. Myo-inositol (mI), often elevated in glial activation, provides additional sensitivity to neuroinflammation and osmoregulation (Coello et al., 2020). Because biochemical alterations may precede morphological change, MRS offers a window into neurochemical imbalance at a stage when sMRI remains normal, making it particularly valuable for monitoring early disease evolution, treatment response, and developmental trajectories (Port, 2020). In this sense, MRS acts as a mechanistic complement to sMRI, DTI, and fMRI by moving beyond anatomy and connectivity to capture underlying metabolic physiology.

Recent advances, most notably MEGA-PRESS editing and ultra-high-field (≥7T) acquisition, have substantially improved spectral separation, enabling more reliable quantification of low-concentration analytes including GABA, glutathione, and lactate (Peek et al., 2023). MRS measures are increasingly integrated within multimodal pipelines and machine-learning frameworks to enhance diagnostic classification, characterize neurochemical signatures of biotypes, and track treatment-linked metabolic normalization (van de Sande et al., 2023). As these applications mature, the technique is shifting from a purely descriptive role toward one of predictive neurochemical profiling within computational psychiatry.

Synthesis of modality-specific findings

Across psychiatric conditions, MRS consistently reveals disturbances in glutamatergic and GABAergic signaling, suggesting a common imbalance in excitatory–inhibitory regulation rather than disorder-specific neurochemical lesions. MDD is frequently associated with reduced Glu and GABA concentrations in prefrontal and anterior cingulate regions, patterns that often normalize with successful antidepressant or neuromodulatory treatment (Hasler et al., 2005; Moriguchi et al., 2019). Schizophrenia tends to show reductions in NAA and altered Glu/Gln cycling within frontal and temporal cortices, consistent with neuronal dysfunction and dysregulated excitatory transmission (Kraguljac et al., 2012; Merritt et al., 2016). ASD frequently demonstrates increased mI and altered Glu–GABA balance in sensory and social–cognition networks, suggesting atypical neurodevelopmental maturation and glial involvement (Brown et al., 2013; Horder et al., 2013). In anxiety disorders, elevated glutamate and reduced inhibitory tone within limbic circuits align with heightened reactivity and impaired emotional regulation (Rosso et al., 2014) (Fig. 4). In bvAD, MRS studies reveal decreased NAA and elevated myo-inositol concentrations in the frontal cortex and posterior cingulate—patterns reflecting neuronal loss and glial activation. Although MRS does not measure tau pathology directly, these metabolic alterations mirror the frontal-dominant tau distribution demonstrated by molecular imaging and are thought to arise from tau-related neurodegeneration. This coupling between cellular metabolic change and frontal tau burden links biochemical disruption to executive and behavioral deficits (Whitwell et al., 2012; Ossenkoppele et al., 2015; Malpetti et al., 2024). Together, these findings highlight MRS as a unique modality linking neurochemical imbalance to network-level dysfunction across psychiatric and neurodegenerative spectra, emphasizing convergent pathways of excitatory–inhibitory dysregulation and glial reactivity.

Figure 4.

MRS abnormalities across six disorders. Spectroscopic profiles summarize neurochemical imbalance across disorders. MDD (1) and bvAD (6) show reduced NAA and elevated choline in frontolimbic regions, schizophrenia (5) shows increased glutamate with reduced GABA, ASD (2) exhibits regionally variable Glu/GABA ratios, OCD (3) shows thalamic choline elevation, and GAD (4) shows increased NAA/Cr. Heatmap intensity reflects reported consistency across studies.

Functional near-infrared spectroscopy

Functional near-infrared spectroscopy (fNIRS) estimates cortical hemodynamic activity by measuring near-infrared light absorption shifts associated with changes in oxygenated and deoxygenated hemoglobin (Li et al., 2023). Owing to its portability, motion tolerance, and patient-friendly design, fNIRS is well suited for populations who struggle with MRI environments—notably children, older adults, and clinically symptomatic groups (Ren et al., 2022). Although its spatial resolution remains lower than that of fMRI, recent hardware and algorithmic advances have expanded cortical coverage and improved temporal resolution. Multichannel arrays now support real-time monitoring of cortical activation during naturalistic behavior, allowing the assessment of cognitive–emotional processing in environments that more closely approximate daily experience.

Increasingly, fNIRS is deployed within multimodal acquisition frameworks, where integration with EEG yields simultaneous electrophysiological and hemodynamic readouts, and where alignment with fMRI strengthens interpretability through cross-validation of connectivity findings. Parallel progress in machine-learning pipelines has facilitated brain-state decoding, classification of psychiatric phenotypes, and early development of closed-loop neurofeedback systems capable of dynamically modulating prefrontal activity (Chang et al., 2021). Emerging clinical applications illustrate growing translational potential: in generalized anxiety disorder, for instance, prefrontal hypoactivation during working-memory and affective challenge tasks reliably correlates with trait anxiety and attentional threat bias (Shen et al., 2025).

Despite these advantages, fNIRS research beyond mood and psychotic disorders remains relatively limited, with studies in ASD, GAD, and related conditions often constrained by small cohorts, heterogeneous task paradigms, and inconsistent preprocessing pipelines. These limitations reduce generalizability and underscore the need for larger, harmonized, and longitudinal datasets, especially those embedded within multimodal frameworks, to determine whether observed patterns represent disorder-specific alterations or reflect broader, transdiagnostic processes.

Synthesis of modality-specific findings

Across conditions, fNIRS most consistently highlights altered prefrontal and frontoparietal activation, mirroring patterns seen in fMRI-derived connectivity and DTI-indexed tract disruption. Depression frequently shows attenuated prefrontal oxygenation during cognitive demand, while anxiety disorders exhibit reduced dorsolateral and ventromedial engagement alongside heightened limbic responsivity, consistent with impaired top-down regulation (Matsuo et al., 2002; Takizawa et al., 2014; Jialin et al., 2025). Preliminary work in autism suggests atypical lateralization and reduced social-cognition-related activation, although findings remain heterogeneous (Kita et al., 2011; Uratani et al., 2019). In OCD, dysfunction within frontal–striatal circuits emerges, aligning closely with corticostriatal abnormalities detected via fMRI and MRS (Mukai et al., 2021; Qiao et al., 2024). In bvAD, functional imaging similarly demonstrates diminished orbitofrontal and medial prefrontal activity corresponding to executive dysfunction and social disinhibition. Although direct fNIRS studies are lacking, this frontal hypometabolism pattern aligns conceptually with reduced prefrontal oxygenation observed in psychiatric populations, reinforcing the role of frontolimbic network failure as a shared substrate of affective and behavioral dysregulation (Whitwell et al., 2012; Ossenkoppele et al., 2015; Lanctot and Fink, 2023). These convergent profiles position fNIRS not as a competing alternative to MRI-based methods but as a complementary, accessible, and ecologically adaptable tool capable of extending multimodal research on prefrontal and salience-network dysfunction across both psychiatric and neurodegenerative conditions.

Positron emission tomography

PET enables molecular-level quantification of neurochemical and physiological processes in vivo, offering a dimension of biological specificity that is not accessible through structural or hemodynamic MRI. Within psychiatric research, its most robust application lies in neuroinflammatory imaging, where translocator protein (TSPO) tracers are used to index microglial activation as a proxy for innate immune activity (Meyer et al., 2020; Malpetti et al., 2024). Building on the limitations of first- and second-generation TSPO ligands, particularly their sensitivity to polymorphic binding affinity, new tracer development has broadened the field considerably. Radioligands targeting monoamine oxidase-B (MAO-B), P2X7 receptors, and colony-stimulating factor-1 receptor (CSF-1R) aim to disentangle astrocytic, microglial, and cytokine-mediated contributions to neuroimmune dysfunction, offering improved mechanistic differentiation (Chen et al., 2021; Cumbers et al., 2024).

Hybrid PET/MRI systems and MRI-based free-water mapping now allow PET signals to be interpreted alongside structural and functional context, reducing ambiguity and improving localization of metabolic or neuroimmune abnormalities (Cheng et al., 2021; Nakaya et al., 2022). Beyond psychiatry, PET maintains an established role in measuring dopaminergic and serotonergic receptor availability and in quantifying glucose metabolism in neurodegenerative disorders such as Alzheimer’s disease, applications that inform comparative interpretation of psychiatric findings (Freiburghaus et al., 2021; Kitamura et al., 2023; Salmon et al., 2024).

Although most psychiatric PET studies remain concentrated in depression, schizophrenia, and bipolar disorder, emerging work is beginning to extend molecular imaging toward neurodevelopmental conditions. A pilot investigation using the mGluR5 ligand [¹⁸F]-FPEB reported increased binding in the postcentral gyrus and cerebellum in adults with ASD, suggesting altered glutamatergic receptor availability in sensorimotor and cerebellar pathways (Fatemi et al., 2018). More recent PET/MRS experiments have replicated mGluR5 abnormalities in small ASD cohorts (Carey et al., 2022). While limited by small sample sizes, male-dominant cohorts, and cross-sectional design, these studies demonstrate both the feasibility and potential of molecular imaging to map receptor-level alterations in ASD and other neurodevelopmental disorders.

Synthesis of modality-specific findings

Across psychiatric conditions, PET consistently reveals alterations in neuroimmune activation and neurotransmitter receptor dynamics, offering mechanistic granularity that is complementary to fMRI connectivity and DTI-based white matter alterations (Ferrando et al., 2025). Depression and schizophrenia frequently show elevated TSPO binding in frontolimbic networks, supporting a model of microglial priming and inflammatory-driven synaptic dysregulation (Setiawan et al., 2015; Bloomfield et al., 2016; Meyer et al., 2020). Dopamine- and serotonin-binding studies further implicate reward, salience, and threat-processing circuits in affective and psychotic illness (Howes and Kapur, 2009). In ASD, preliminary mGluR5 findings point to glutamatergic imbalance in sensory-motor and cerebellar pathways, mirroring spectroscopic reports of altered Glu–GABA equilibrium (Ferrando et al., 2025; Naples et al., 2026). In bvAD, fluorodeoxyglucose (FDG)-PET reveals a characteristic pattern of hypometabolism involving the orbitofrontal, medial prefrontal, and anterior cingulate cortices, typically accompanied by reduced activity in posterior temporoparietal regions. This dual frontal–posterior signature differentiates bvAD from frontotemporal dementia while explaining its prominent behavioral and dysexecutive presentation. Amyloid PET consistently demonstrates diffuse cortical β-amyloid deposition, confirming Alzheimer pathology despite the atypical clinical phenotype, while tau PET highlights a frontal-predominant distribution of neurofibrillary burden that correlates closely with executive dysfunction and neuropsychiatric symptoms. These findings underscore how bvAD bridges psychiatric and neurodegenerative domains, with molecular PET biomarkers clarifying the mechanistic underpinnings of frontolimbic network degeneration (Rabinovici et al., 2011; Ossenkoppele et al., 2015; Malpetti et al., 2024). Together, PET findings across psychiatric and neurodegenerative spectrums support a unified model in which neuroimmune activation, receptor dysregulation, and metabolic compromise converge on frontolimbic and salience networks, linking molecular pathology with behavioral expression.

In summary, Table 1 synthesizes each modality’s key strengths, limitations, and clinical applications, while Table 2 summarizes representative cross-disorder findings spanning all six techniques discussed above. Alongside the tables, Figs 1–4 visually depict cross-disorder alterations derived from structural, diffusion, functional, and spectroscopic modalities. The color intensity reflects the relative direction and strength of inter-regional associations summarized from the literature, normalized between –1 and +1 for cross-modal comparison. Warmer tones indicate stronger positive associations such as higher co-atrophy or increased connectivity, whereas cooler tones indicate reduced connectivity or opposing alterations. These maps synthesize findings from prior meta-analytic and consortium studies and should be interpreted as illustrating relative patterns of convergence and directionality across disorders rather than exact statistical effect sizes.

Table 1.

Summary of primary neuroimaging modalities in psychiatric and cognitive disorders. The summaries reflect representative findings from the literature and are not exhaustive or meta-analytic.

Modality	Measures	Strengths	Limitations	Common psychiatric applications
sMRI	Cortical thickness, brain volume, surface area	High spatial resolution, well-validated, suitable for longitudinal studies	Low sensitivity to dynamic physiological processes	Morphometric analysis, disease progression, subtyping
DTI	White matter integrity (FA, MD, RD, AD)	Sensitive to microstructural changes, supports connectome analysis	Motion-sensitive, low specificity, partial volume effects	White matter tract pathology, connectivity analysis
fMRI	Neural activity via BOLD signal	Captures resting-state and task-related brain networks, rich temporal data	Complex preprocessing, susceptible to motion and physiological noise	Network dysfunction, emotion and cognitive control, prediction models
MRS	Neurochemicals (e.g. NAA, Glu, GABA, Cho, Cr, mI)	Non-invasive neurochemical profiling, repeatable	Low spatial resolution, technical variability	Excitatory/inhibitory balance, glial markers, treatment monitoring
fNIRS	Hemodynamic responses (HbO/HbR)	Portable, motion-tolerant, child-friendly, real-time capable	Limited depth and spatial resolution, mostly cortical	Task-based cortical activation, emotion and attention monitoring
PET	TSPO binding, neurotransmitters, glucose metabolism	Molecular specificity, targets inflammation and receptor dynamics	Radiation exposure, high cost, limited availability	Microglial activation, neuroinflammatory and hypometabolic markers

AD—axial diffusivity; BOLD—blood-oxygen-level dependent; Cho—choline; Cr—creatine; DTI—diffusion tensor imaging; FA—fractional anisotropy; fMRI—functional magnetic resonance imaging; fNIRS—functional near-infrared spectroscopy; GABA—gamma-aminobutyric acid; Glu—glutamate; HbO—oxygenated hemoglobin; HbR—deoxygenated hemoglobin; MD—mean diffusivity; mI—myo-inositol; MRS—magnetic resonance spectroscopy; NAA—N-acetylaspartate; PET—positron emission tomography; RD—radial diffusivity; sMRI—structural magnetic resonance imaging; TSPO—translocator protein.

Table 2.

Salient neuroradiological findings across major psychiatric and cognitive disorders. The summaries reflect representative findings from the literature and are not exhaustive or meta-analytic. Representative references correspond to key multimodal or cross-disorder studies and are provided as exemplary sources rather than a comprehensive bibliography.

Disorder	sMRI	DTI	fMRI	MRS	fNIRS	PET	Representative references
MDD	↓ Hippocampus, PFC, ACC, amygdala volume	↓ FA in uncinate fasciculus, corpus callosum, cingulum	DMN hyperconnectivity; ↓ FPN connectivity	↓ NAA, ↑ Cho in PFC	↓ PFC activation during cognitive tasks	↑ FW in WM; ↓ MTR	(Matsuo et al., 2002; Liao et al., 2013; Bessette et al., 2014; Setiawan et al., 2015; Schmaal et al., 2017; Moriguchi et al., 2019; Coello et al., 2020; McTeague et al., 2020; Meyer et al., 2020; Opel et al., 2020; Van Velzen et al., 2020; Wang et al., 2024; Wang et al., 2025)
ASD	Early overgrowth in frontal & temporal lobes	↓ FA in corpus callosum, arcuate fasciculus, cingulum	↓ Long-range & ↑ local connectivity (DMN, social)	↓ NAA, GABA; ↑ Glu in PFC	↓ Frontal activation in social/emotional tasks	Limited PET data; ↑ cytokine evidence from peripheral studies	(Kita et al., 2011; Travers et al., 2012; Brown et al., 2013; Horder et al., 2013; Uddin et al., 2013; Ameis and Catani, 2015; Ecker et al., 2015; Fornito et al., 2015; Hazlett et al., 2017; Hull et al., 2017; Fatemi et al., 2018; Uratani et al., 2019; Carey et al., 2022; Chien et al., 2024)
OCD	↑/↓ Volume in OFC, ACC, basal ganglia	↓ FA in frontostriatal & thalamocortical pathways	Hyperactive CSTC loop (OFC, caudate)	↓ tCr, mI; ↑ Cho in thalamus	Understudied	↑ FW, ↑ TSPO binding	(Harrison et al., 2009; Beucke et al., 2013; Piras et al., 2013; McTeague et al., 2017; Boedhoe et al., 2020; Meyer et al., 2020; Opel et al., 2020; Biria et al., 2021; Mukai et al., 2021; Qiao et al., 2024)
GAD	↓ GM in amygdala, mPFC, insula (variable)	↓ FA in uncinate fasciculus, cingulum	↑ Amygdala/insula activity; ↓ PFC–amygdala coupling	↑ NAA/Cr ratio in PFC	↓ PFC response in affective tasks	Limited PET data; ↑ cytokine evidence from peripheral studies	(Trzesniak et al., 2008; Sylvester et al., 2012; Tromp et al., 2012; Najjar et al., 2013; Rosso et al., 2014; Wang et al., 2016; McTeague et al., 2017; Wang et al., 2018; Meyer et al., 2020; Jialin et al., 2025; Shen et al., 2025)
SCZ	↓ GM in STG, thalamus, hippocampus, PFC	↓ FA in frontotemporal tracts, internal capsule, cingulum	Hypoconnectivity in DMN, ECN; SN dysregulation	↑ Glu, ↓ GABA in PFC; NMDA hypofunction	↓ PFC activity in executive tasks	↑ FW; ↑ microglial activation via TSPO PET	(Kraguljac et al., 2012; Woodward et al., 2012; Wheeler and Voineskos, 2014; Fornito et al., 2015; Merritt et al., 2016; McTeague et al., 2017; Kelly et al., 2018; Van Erp et al., 2018; Meyer et al., 2020; Opel et al., 2020)
bvAD	↓ GM in frontal cortex, medial temporal lobe, PCC; atrophy in salience and limbic areas	↓ FA in parahippocampal WM, cingulum	↓ DMN coherence; disrupted salience and social cognition networks	↓ NAA, ↑ mI in frontal cortex and PCC	↓ PFC oxygenation during emotion regulation tasks	Frontal-predominant ↓ glucose metabolism on FDG PET; TSPO PET data lacking	(Fornito et al., 2015; Ossenkoppele et al., 2015; Singleton et al., 2020; Lanctot and Fink, 2023; Malpetti et al., 2024)

ACC—anterior cingulate cortex; ASD—autism spectrum disorder; bvAD—behavioral variant of Alzheimer’s disease; Cho—choline; Cr—creatine; CSTC—cortico-striato-thalamo-cortical circuit; DMN—default mode network; DTI—diffusion tensor imaging; ECN—executive control network; FA—fractional anisotropy; FDG—fluorodeoxyglucose; fMRI—functional magnetic resonance imaging; fNIRS—functional near-infrared spectroscopy; FPN—frontoparietal network; FW—free water; GABA—gamma-aminobutyric acid; GAD—generalized anxiety disorder; Glu—glutamate; GM—gray matter; MDD—major depressive disorder; mI—myo-inositol; mPFC—medial prefrontal cortex; MRI—magnetic resonance imaging; MRS—magnetic resonance spectroscopy; MTR—magnetization transfer ratio; NAA—N-acetylaspartate; NMDA—N-methyl-d-aspartate; OCD—obsessive-compulsive disorder; OFC—orbitofrontal cortex; PCC—posterior cingulate cortex; PFC—prefrontal cortex; PET—positron emission tomography; SCZ—schizophrenia; sMRI—structural magnetic resonance imaging; SN—salience network; STG—superior temporal gyrus; tCr—total creatine; TSPO—translocator protein; WM—white matter

Cross-modal integration and mechanistic convergence

Advances in analytic frameworks now enable the direct integration of structural, functional, and molecular neuroimaging data within unified models of brain organization. Multimodal data fusion approaches such as joint independent component analysis and linked matrix factorization extract latent components that capture shared variance across modalities, linking structure, connectivity, and behavior (Groves et al., 2011; Calhoun and Sui, 2016). Deep-learning architectures, including convolutional and graph neural networks, extend this principle by integrating MRI, PET, and clinical data to predict diagnostic status, symptom severity, and treatment response (Vieira et al., 2017; Du et al., 2018). These developments provide the empirical foundation for examining cross-modal convergence in psychiatric neuroimaging.

Convergent evidence from DTI, fMRI, and MRS supports a systems-level interpretation, in which psychiatric syndromes reflect distributed network dysfunction rather than isolated regional deficits. The overlap between psychiatric and neurodegenerative conditions, particularly bvAD, suggests that diverse upstream mechanisms, whether developmental, inflammatory, or degenerative, can produce similar network-level collapse. Mechanistically, these circuits occupy metabolically demanding, transcriptionally diverse cortical territories, explaining their heightened vulnerability to inflammatory and stress-related insults (Fornito et al., 2015; Burt et al., 2018; Vázquez-Rodríguez et al., 2019).

The integration of multimodal findings further highlights the need for a systematic framework for biomarker discovery. Figure 5 summarizes this workflow, from cohort selection through multimodal acquisition, data fusion, statistical modeling, and clinical interpretation. This pipeline bridges disorder-specific results with broader translational objectives, illustrating how cross-modal integration and cross-diagnostic comparison support the development of clinically meaningful biomarkers. While convergent evidence suggests shared structural and functional alterations across psychiatric disorders, interpretation must remain cautious. Effect sizes are modest and replication inconsistent, and methodological variability limits comparability. Publication bias and analytic heterogeneity further risk overstating cross-study agreement. Many similarities may reflect shared confounders such as medication exposure, smoking, sleep disruption, or comorbidity rather than true mechanistic overlap. Reverse inference adds uncertainty, as similar imaging patterns may arise from distinct biological pathways. Thus, neuroimaging convergence should be viewed as supportive but not definitive evidence of shared pathology.

Figure 5.

Conceptual workflow in neuroradiology of psychiatric disorders. Schematic representation of the biomarker discovery process, outlining participant selection, multimodal imaging acquisition, data integration, statistical modeling, and clinical interpretation. The figure provides an overview linking the reviewed findings to translational biomarker development.

Biological mechanisms contributing to cross-modal convergence: neuroinflammatory, neurovascular, and glymphatic pathways

Emerging data implicate neuroinflammatory, neurovascular, and glymphatic dysfunction in psychiatric disorders. Neuroinflammation and blood–brain barrier (BBB) dysfunction are increasingly recognized as central mechanisms in the pathophysiology of psychiatric illness. MDD, schizophrenia, bipolar disorder, and ASD show accumulating evidence of neuroimmune involvement, reflecting interactions between inflammatory cascades, vascular integrity, and neural signaling (Najjar et al., 2013). Elevated cytokines such as IL-6 and TNF-α, persistent low-grade inflammation, and microglial activation compromise BBB function, allowing peripheral immune mediators to penetrate the central nervous system and disrupt astrocytic regulation (Huang et al., 2021). MRI-based permeability mapping, including dynamic contrast-enhanced imaging (DCE-MRI) and diffusion-derived permeability indices, has demonstrated increased BBB leakage in schizophrenia and MDD, particularly within the hippocampus and prefrontal cortex, regions integral to emotion regulation and executive control (Pollak et al., 2018). Notably, depression subtypes with elevated inflammatory markers tend to show more severe symptoms and reduced treatment responsiveness (Yang et al., 2019), suggesting potential stratification value.

Emerging PET tracers targeting neuroimmune pathways including TSPO, MAO-B, P2X7, and CSF-1R offer molecular-level resolution of inflammatory processes, while diffusion-prepared arterial spin labeling (ASL) methods (e.g. K-trans, the volume transfer constant, and KW, reflecting water exchange across BBB) provide indirect assessment of permeability and perfusion (Setiawan et al., 2015). When integrated with fMRI or network-level analyses, these biomarkers suggest that functional dysconnectivity may reflect downstream consequences of immune-mediated neurovascular instability rather than isolated synaptic disruption. Increasingly, vascular alterations appear to function not merely as correlates of inflammation, but as the structural interface through which immune dysregulation exerts neural effects.

Within this framework, neurovascular integrity has emerged as a mechanistic extension of immune dysfunction. Dynamic contrast-enhanced (DCE)-MRI studies report subtle but regionally specific BBB permeability increases in MDD and schizophrenia across the hippocampus, anterior cingulate, and prefrontal cortex (Shang et al., 2024). Pharmacokinetic modeling of K-trans demonstrates strong associations between BBB permeability and symptom severity (Voorter et al., 2024), and cortical thinning in areas with BBB leakage further suggests that microvascular dysfunction may precede structural degeneration (Zhang et al., 2022). DTI studies support this interpretation, showing reductions in fractional anisotropy and increased mean diffusivity within frontolimbic tracts including the uncinate fasciculus and cingulum in settings of increased BBB permeability (Elschot et al., 2021). These patterns probably reflect inflammatory-driven astrocytic swelling, perivascular edema, and secondary white matter injury (Raja et al., 2018).

A third dimension, glymphatic clearance dysfunction, has gained prominence as a convergent mechanism linking neuroinflammation, sleep disruption, and cognitive impairment. The glymphatic system facilitates metabolic waste transport along perivascular pathways, and its impairment has now been demonstrated in psychiatric conditions using diffusion tensor imaging analysis along the perivascular space (DTI-ALPS). Patients with depression show significantly reduced ALPS indices, correlating with higher symptom burden, fatigue scores, and microstructural white matter abnormalities in emotional-regulation tracts (Yang et al., 2024; Bao et al., 2025). Increased choroid plexus volume—interpreted as a proxy of immune activation and altered cerebrospinal fluid filtration—frequently co-occurs with reduced glymphatic efficiency, and enlargement of Virchow–Robin spaces demonstrates further disruption of perivascular flow (Xu et al., 2024; Yao et al., 2024). These findings suggest that depression may, at least in part, reflect impaired neurofluid transport and perivascular homeostasis, integrating inflammatory, vascular, and clearance mechanisms into a unified pathobiological model.

The shared presence of glymphatic failure in both depression and Alzheimer’s disease highlights continuity between psychiatric and neurodegenerative spectra, reinforcing the notion that waste-clearance inefficiency and immune–vascular interaction contribute to long-term network destabilization. When viewed alongside the structural, functional, and microstructural signatures reviewed earlier, these multimodal findings converge on a mechanistic model in which immune dysregulation, BBB instability, and impaired clearance form a physiologically coherent substrate underlying variable clinical phenotypes.

In summary, neuroinflammation, BBB permeability, and glymphatic disturbance reflect interdependent, transdiagnostic pathways rather than isolated disease-specific abnormalities. Developmental timing, genetic predisposition, environmental exposures, and compensatory plasticity probably determine whether shared mechanisms are expressed as mood disorder, psychosis, or neurodegeneration. Growing evidence supports classifying psychiatric illness using neural phenotypes and biomarker constellations rather than symptom-based nosology (Patel et al., 2021), underscoring the value of multimodal imaging in defining biological dimensions of disease.

Translational and therapeutic implications

Neuropsychiatric disorders such as MDD, schizophrenia, and OCD are increasingly understood as network-based conditions, motivating circuit-targeted therapies. Transcranial magnetic stimulation (TMS) combined with fMRI enables individualized neuromodulation by mapping dysfunctional networks and tailoring stimulation targets (Baliga and Mehta, 2021). In MDD, dorsolateral prefrontal stimulation is guided by its anticorrelation with the subgenual anterior cingulate cortex; in OCD, fronto-striatal and SMA-focused stimulation reduces compulsivity and normalizes hyperconnectivity (Reddy et al., 2023; Oathes et al., 2024). In schizophrenia, interventions seek to restore fronto-temporal integration, while in GAD and PTSD, amygdala–prefrontal modulation targets salience and fear circuitry (Patel et al., 2023).

Pharmacologic treatment also reshapes connectivity. Antipsychotic-linked modulation of DMN, salience, and frontoparietal networks correlates with clinical improvement (Kraguljac et al., 2016; Wang et al., 2017; Sreeraj et al., 2023). However, despite the promise of imaging-guided interventions, few biomarkers meet clinical-translation criteria including reliability, sensitivity, specificity, and real-world utility (Etkin and Mathalon, 2024). Variability in imaging pipelines and sample heterogeneity remains a major barrier. Normative modeling, federated analytics, and large-scale initiatives such as ENIGMA and COINSTAC offer pathways toward reproducible, translatable biomarkers. Integration of these frameworks with circuit-informed neuromodulation will be critical for next-generation precision therapies.

Limitations and future directions

Diagnostic nosology

Psychiatric diagnoses remain broad, overlapping, and symptom-based, producing substantial heterogeneity in imaging results (Abi‐Dargham et al., 2023). Shared patterns may arise from distinct etiologies converging on similar circuits, complicating disorder-specific biomarker identification. Future studies must incorporate longitudinal, medication-naïve cohorts and integrate genetic, inflammatory, and behavioral metrics to distinguish causal pathology from secondary or compensatory effects.

Methodological standardization

Reproducibility is hindered by variable acquisition, sample size, statistical pipelines, and cross-sectional designs. Group-level effects are often modest and rarely predictive at the individual scale. Confounders—including medication status, lifestyle factors, and demographic diversity—further erode signal consistency (Calhoun et al., 2021). Standardized imaging protocols, harmonized preprocessing, and longitudinal designs are needed. Large collaborative datasets (ENIGMA, UK Biobank) are accelerating this process.

Translational barriers

Most biomarkers remain at the research stage, lacking independent replication or demonstrated clinical utility. Future progress requires validation pipelines, integration with multimodal biological data, and alignment with actionable diagnostic or treatment endpoints. Emerging computational approaches, including normative modeling and cross-site federated analysis, show promise for mapping biologically meaningful subtypes across disorders (Marquand et al., 2016; Sui et al., 2020; Liloia et al., 2024). The next critical step is to link these biomarker frameworks to real clinical decision-making.

Conclusion

This review provides an integrated, cross-modal comparison of neuroimaging signatures in psychiatric illness, offering a unified framework that contrasts findings across structural, metabolic, connectivity-based, molecular, and neurovascular modalities. By moving beyond disorder-specific summaries to transdiagnostic synthesis, we outline a biologically grounded perspective that complements emerging models in psychiatric neuroscience. Although progress has been significant, clinical translation demands rigorous standardization, multimodal integration, and biomarker validation at the individual level. Achieving these aims will enable neuroradiology to contribute meaningfully to precision psychiatry.

Author contributions

Sina Dindarian (Conceptualization, Data curation, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing), Joga Chaganti (Data curation, Investigation, Methodology, Supervision, Visualization, Writing – original draft, Writing – review & editing), Nazanin Rafiei (Data curation, Investigation, Writing – original draft, Writing – review & editing), Scott Faro (Investigation, Project administration, Resources, Supervision, Writing – review & editing).

Conflicts of interests

The authors declare no conflict of interests.