Functional Near-Infrared Spectroscopy in Schizophrenia Research: Progress, Challenges, and Future Directions

Abstract

This review comprehensively examines the application and progress of functional near-infrared spectroscopy (fNIRS) in schizophrenia research. Schizophrenia is a complex neuropsychiatric disorder characterized by extensive dysfunction in the prefrontal-limbic system and dysregulation of brain network connectivity. fNIRS, with its advantages of high portability, resistance to motion interference, and non-invasive real-time monitoring of cerebral hemodynamic responses, has emerged as a valuable tool in exploring the neural mechanisms of schizophrenia. It can capture the overactivity of the default mode network (DMN) during resting states and the under-activation of brain regions such as the prefrontal and temporal lobes during task states. Studies have demonstrated that combining fNIRS with neuromodulation techniques, including transcranial direct current stimulation (tDCS) and theta burst stimulation (TBS), can improve negative symptoms and offer a potential approach for individualized diagnosis and treatment. However, challenges remain in fNIRS research, such as signal noise processing, insufficient ability to detect deep brain regions, and research heterogeneity. Future research will focus on technological innovations (eg, high-density fNIRS and multimodal fusion), machine-learning-driven biomarker mining, and clinical translation (eg, optimizing individualized neuroregulatory targets). Through interdisciplinary integration, fNIRS is expected to facilitate breakthroughs in the individualized diagnosis and treatment of schizophrenia and enhance understanding of its neural mechanisms. This review also discusses the potential of fNIRS as a biomarker for schizophrenia, its role in monitoring treatment efficacy, and the importance of addressing technical limitations to advance its clinical application.

Introduction

Schizophrenia is a complex neurodevelopmental disorder, which has been confirmed by several studies to involve extensive prefrontal-limbic system dysfunction and brain network connectivity dysregulation [1,2]. Traditional neuroimaging methods like functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) have contributed to our understanding of schizophrenia; however, they have limitations such as high cost, susceptibility to motion artifacts, and poor temporal resolution. Schizophrenia patients often cannot tolerate fMRI/electrocardiogram (EEG) constraints during psychotic states. These constraints may hinder the comprehensive exploration of the dynamic neural mechanisms underlying schizophrenia. fNIRS compensates for the lack of spatial resolution of EEG. The mean spatial error of EEG is ±8.2 mm, and the error of functional near-infrared spectroscopy (fNIRS) is ±3.5mm, which confirms its advantage in cortical localization [3]. EEG is easily disturbed by muscle activity and head movement, but fNIRS is more tolerant to motion artifacts and is more suitable for use in dynamic environments [4,5].

The advent of fNIRS has opened new avenues for schizophrenia research. fNIRS is a non-invasive neuroimaging technique that measures changes in the concentration of oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) in the brain cortex. It infers local hemodynamic responses triggered by neural activity based on the differing light absorption spectra of HbO and HbR [6]. This technique offers several advantages, such as high portability, resistance to motion interference, and real-time monitoring capabilities, making it particularly suitable for studying schizophrenia patients, including those with motor restlessness [7]. Hemodynamic changes are highly relevant to understanding schizophrenia. Neural activity is accompanied by changes in cerebral blood flow and blood oxygenation levels. In schizophrenia patients, abnormal neural activity leads to altered hemodynamic responses, which can be measured by fNIRS. These hemodynamic changes may serve as potential biomarkers for schizophrenia diagnosis and treatment efficacy evaluation.

Studies have shown that fNIRS can accurately detect changes in oxygenated hemoglobin (HbO) concentration in the prefrontal cortex and reveal brain activation patterns during cognitive tasks [8]. In addition, the integration of fNIRS with neuromodulation techniques opens new avenues for real-time monitoring of intervention effects and the construction of closed-loop rehabilitation strategies [9,10]. fNIRS can detect abnormalities in resting-state brain networks [11], and higher cognitive function deficits, and evaluate neuroplasticity, providing a unique perspective to reveal the neural mechanisms of schizophrenia [12]. Future studies are needed to further integrate AI with multimodal imaging techniques. None of the current reviews synthesizes the progress of fNIRS in neuromodulation monitoring in schizophrenia, such as TBS/tDCS, and machine-learning biomarker discovery that is critical for individualized intervention. This article reviews the above studies and discusses future prospects to promote the development of individualized diagnosis and treatment strategies.

Overview of fNIRS Technology

Technical Principles

The Relationship Between Near-Infrared Light and Cerebral Hemodynamics

fNIRS emits near-infrared light (wavelength 650–950 nm) to penetrate the scalp and skull, detecting changes in HbO and deoxygenated hemoglobin (HbR) concentrations in brain tissue, thereby indirectly reflecting local hemodynamic responses triggered by neural activity [13]. Based on the differing light absorption spectra of HbO and HbR, multi-wavelength measurements enable dynamic separation of them [14]. This non-invasive method is particularly suitable for real-time monitoring of cerebral cortical hemodynamic activity [15]. The principle of fNIRS imaging is shown in Figure 1.

Figure 1. Schematic diagram of fNIRS imaging.

Panel A illustrates the wavelength dependent absorption of hemoglobin and oxygen and hemoglobin, highlighting the utility of the NIR window in reducing tissue attenuation. Panel B depicts a cross-sectional view of the human head, showing light propagation through the anatomical layers and detector placement. This equation formalizes the relationship between changes in optical density and hemoglobin dynamics.

Spatial and Temporal Resolution Characteristics

The spatial resolution of fNIRS (approximately 1–3 cm) is limited by the distance between the light source and detector, and it is sensitive to superficial cortical areas (such as the prefrontal cortex) but has limited ability to capture signals from deep brain regions [16]. Its temporal resolution (approximately 0.1–10 Hz) is significantly better than functional magnetic resonance imaging (fMRI), allowing the capture of rapid hemodynamic fluctuations, but is lower than the millisecond precision of electroencephalography (EEG) [17]. Additionally, motion artifacts and environmental interference are major sources of noise in fNIRS, requiring hardware improvements (such as flexible probes) and algorithm optimization (such as wavelet decomposition) to enhance signal quality [14,18].

Comparison of fNIRS with Other Neuroimaging Techniques

Comparison with fMRI and EEG

Compared to fMRI, fNIRS offers better portability, cost-effectiveness, and resistance to motion artifacts, making it suitable for dynamic tasks in natural settings, but it has lower spatial resolution and cannot detect whole-brain activity [15,17]. Compared to EEG, fNIRS measures hemodynamic responses directly rather than electrophysiological activity, avoiding muscle artifact interference and insensitivity to scalp impedance changes, but it has lower temporal resolution and cannot capture instantaneous characteristics of neural oscillations [18]. The specific comparison is shown in Table 1.

Table 1.

Comparison of fNIRS, fMRI, and EEG across key parameters.

Parameter	fNIRS	fMRI	EEG
Spatial resolution	Moderate (1–3 cm)	High (mm to sub-mm)	Low (source localization challenges)
Temporal resolution	High (0.1–10 Hz)	Moderate (seconds)	Very high (millisecond range)
Portability	High	Low	Moderate
Motion tolerance	High	Low	Moderate
Cost	Low to moderate	High	Low
Suitability for schizophrenia research	Ideal for dynamic tasks and natural settings, limited deep brain detection	Excellent for whole-brain imaging, limited portability	Excellent for real-time electrophysiological activity, limited spatial precision

Table 1 compares functional near-infrared spectroscopy (fNIRS), functional magnetic resonance imaging (fMRI), and electroencephalography (EEG) across 6 key neuroimaging parameters relevant to schizophrenia research. Parameters include spatial/temporal resolution, portability, motion tolerance, cost, and modality-specific suitability for studying schizophrenia. fNIRS balances portability and motion tolerance with moderate spatial resolution, while fMRI offers superior spatial resolution at the expense of portability. EEG excels in temporal resolution but has spatial localization limitations.

Applicable Scenarios in Schizophrenia Research

fNIRS can detect abnormalities in prefrontal cortex (PFC) hemodynamics, such as reduced HbO signals during tasks or compensatory activation, which are related to negative symptoms [19]. It is suitable for cognitive tasks requiring natural interaction (such as verbal fluency tests), avoiding the claustrophobic limitations of fMRI [13]. fNIRS can also monitor brain function in daily settings over long periods, supporting the tracking of daily brain function in schizophrenia patients and evaluating the efficacy of medications or interventions [17]. However, its insufficient spatial resolution may limit the analysis of complex neural networks, necessitating the integration of multimodal imaging to enhance interpretability [16].

Applications of fNIRS in Schizophrenia Research

Resting-State Studies

Abnormal Functional Connectivity in Key Brain Regions Such as the PFC

fNIRS studies have revealed abnormal functional connectivity in the PFC of schizophrenia patients. For example, Akın et al analyzed the neurocognitive ratio (NCR) using fNIRS technology and found significantly reduced functional connectivity in the PFC region of schizophrenia patients [20]. Using principal component analysis (PCA) to separate the cognitive mode network (CMN) from the default mode network (DMN), it was found that the efficiency of the CMN in schizophrenia patients was significantly lower than in healthy controls, indicating impaired functional coordination in the prefrontal-subcortical circuit. Studies also found that the DMN activity in schizophrenia patients was excessively active during resting states, and the switching efficiency between the DMN and task-related networks was reduced. This “disinhibited” state of the DMN may reflect the pathological characteristics of excessive intrinsic thought activity and insufficient external attention regulation in patients. Additionally, Cattarinussi et al’s review of the neural mechanisms of catatonia pointed out that abnormal interactions between the DMN and motor control networks may be widespread in the schizophrenia spectrum [21].

Collectively, resting-state fNIRS studies consistently demonstrate prefrontal functional connectivity abnormalities in schizophrenia, characterized by reduced cognitive mode network (CMN) efficiency and DMN hyperactivation. This DMN “disinhibition” likely reflects pathological over-engagement in internal mentation and impaired attentional switching. Critically, aberrant DMN-motor network interactions provide a neural framework for catatonia across the schizophrenia spectrum [20,22].

Task-State Studies

Cerebral Hemodynamic Changes in Cognitive Function Tasks

In working memory tasks, Kumar et al found that patients exhibited significantly reduced HbR concentrations in the right frontal pole compared to healthy controls when performing high-load tasks (such as 2-back) and that the equivalent dose of olanzapine was negatively correlated with activation in the right frontal pole, suggesting that antipsychotic medications may suppress compensatory hyperactivation [18]. Furthermore, Ma et al validated the specific activation of the left dorsolateral prefrontal cortex (DLPFC) during the 2-back task using fNIRS, providing a neurophysiological basis for subsequent “functionally targeted” treatments with transcranial direct current stimulation (tDCS) [23]. In the verbal fluency task (VFT), studies on clinically high-risk (CHR) individuals showed that the HbO levels in the right superior temporal gyrus (rSTG) during the VFT were significantly lower than in healthy controls, and the degree of activation was positively correlated with working memory performance [24]. This suggests that temporal lobe dysfunction may be a neural marker of early psychosis. In terms of the association between genetics and cognitive function, Kopf et al found that the NOS1 gene ex1f-VNTR polymorphism affects prefrontal oxygenation levels during working memory tasks, with carriers of the short allele exhibiting higher DLPFC activation, possibly compensating for neurofunctional deficits related to the gene [25]. This provides a new perspective on the genetic-cognitive interaction mechanisms in schizophrenia.

In summary, task-based fNIRS reveals distinct neurofunctional signatures: 1) Compensatory right frontopolar hyperactivation during high-load working memory (suppressed by antipsychotics) [22]; 2) Left DLPFC specificity during 2-back tasks, validating it as a neuromodulation target [23]; 3) Right STG hypoactivation in CHR individuals as an early psychosis biomarker linked to working memory deficits [24]; and 4) NOS1 gene polymorphism effects on prefrontal oxygenation, highlighting gene-cognition interplay [25].

Abnormal Activation Patterns in Social Cognition and Emotion Processing Tasks

Deficits in social cognition are a core characteristic of schizophrenia, and fNIRS studies have revealed unique neural patterns in patients during tasks such as emotion recognition and theory of mind (ToM). In an emotional facial recognition task, Hirata et al compared brain activity differences between autism spectrum disorder (ASD) and schizophrenia patients, finding that activation in the left prefrontal-temporal region during the emotional facial recognition task was significantly lower in schizophrenia patients than in healthy controls but higher than in ASD, suggesting that the 2 disorders may have different neuropathological mechanisms [26]. Sayar-Akaslan et al found through a “mind-reading task” that activation in the premotor area, superior temporal gyrus, and primary somatosensory cortex during ToM tasks was significantly lower in schizophrenia and bipolar disorder patients than in healthy controls, and the activation patterns between patient groups were similar, supporting partial overlap in the neural basis of social cognition across mental disorders [27]. Singh et al, combining multimodal data (fNIRS, EEG, and facial behavior), found that early schizophrenia patients exhibited abnormal functional connectivity in the prefrontal-temporal network during face-to-face social interactions, and geometric features (such as curvature and path features) could effectively distinguish patients from controls, providing new evidence for the dynamic neural mechanisms of social cognitive deficits [28].

Three critical patterns emerge: 1) Schizophrenia shows intermediate prefrontal-temporal hypoactivation (between ASD and HC) during emotion processing, suggesting disorder-specific pathology [26]; 2) Premotor-temporal-somatosensory hypoactivation during ToM tasks reveals shared neural deficits with bipolar disorder [27]; and 3) Multimodal approaches (fNIRS-EEG-behavioral) expose dynamic prefrontal-temporal dysconnectivity during social interactions, enabling geometric feature-based classification [28].

fNIRS task-state studies have revealed abnormal activation patterns in key brain regions such as the prefrontal and temporal cortices in schizophrenia patients. Future research needs to further explore the standardization of task design, longitudinal neuroplasticity changes, and optimization of individualized treatment targets [23,29].

Symptom-Related Analysis

Correlation Between Positive Symptoms and Specific Brain Region Activity

To date, fNIRS-based studies have paid relatively little attention to the neural mechanisms of positive symptoms in schizophrenia, but some studies indirectly suggest that dysfunction in the temporal and prefrontal regions may be related. For example, Wei et al found that in clinically high-risk (CHR) individuals performing the VFT, HbO activation in the rSTG was significantly reduced, and this hypoactivation was associated with working memory deficits [24]. Since CHR states are often accompanied by mild positive symptoms, this finding suggests that temporal lobe dysfunction is related to the early neural mechanisms of positive symptoms. Additionally, Curtin et al found in a visuospatial attention task that activity in the left middle frontal gyrus (MFG) of schizophrenia patients was significantly correlated with negative symptoms [30]. Combining other imaging studies, it is speculated that positive symptoms may involve dysfunction in the temporal-frontal network, but more fNIRS studies are needed to clarify their specific associations. While most studies reported prefrontal hypoactivation, Kumar et al [22] observed compensatory hyperactivation in the right frontal pole during high-load working memory tasks potentially masked by antipsychotic suppression. This highlights task-load and medication status as critical interpretation confounders.

Neurobiological Markers of Negative Symptoms

fNIRS studies consistently indicate that negative symptoms are closely related to hypoactivation or abnormal functional connectivity in the PFC. Ma et al found that in schizophrenia patients with negative symptoms, activation in the left DLPFC during the 2-back working memory task was significantly enhanced, while the VFT did not elicit activity in this region [23]. This suggests that the functional plasticity of the left DLPFC may serve as a target for neuromodulation techniques such as tDCS. Gao et al, in a θ-burst stimulation (TBS) intervention study, found that stimulating the bilateral DMPFC significantly improved negative symptoms and enhanced prefrontal HbO activation and the small-world properties of functional networks. TBS improves negative symptoms by 40% with concurrent fNIRS-verified prefrontal normalization [31]. These changes were positively correlated with symptom improvement, suggesting that functional recovery of the DMPFC may alleviate negative symptoms by enhancing the information integration efficiency of the prefrontal-subcortical circuit. Curtin et al further found that abnormal activation in the left MFG during visuospatial tasks was significantly correlated with PANSS scores for negative symptoms, indicating that dysfunction in this region during attention regulation may directly drive core negative symptoms such as emotional blunting [30].

fNIRS studies have revealed that negative symptoms are closely related to functional abnormalities in multiple prefrontal subregions (DLPFC, DMPFC, MFG), while positive symptoms may involve early functional damage in the temporal lobe (eg, TG). Neuromodulation techniques show potential for improving negative symptoms. However, current research still has limitations: there is insufficient fNIRS evidence for positive symptoms, the impact of task design (such as n-back and VFT) on activation patterns needs further standardization, and sample sizes were generally small.

Monitoring of Treatment and Intervention

Effects of Antipsychotic Medications on Cerebral Hemodynamics

Using an n-back working memory task, Kumar et al found a 20–30% reduction in HbO activation in the DLPFC and an inverse correlation between equivalent doses of olanzapine and HbR concentration in the right frontal pole (BA10), suggesting that the drug improves cognitive function by inhibiting excessive activation in this region [22]. However, this inhibitory effect may be accompanied by overall prefrontal hypoactivation, as shown by Bhargav et al, who found that patients treated with atypical antipsychotics had significantly lower activation of HbO and total hemoglobin in the bilateral prefrontal cortex during yoga breathing training compared to healthy controls, indicating that the medication partially limits neuroplasticity [32]. Additionally, Iwashiro et al found that the correlation between gray matter volume and functional activity in the Broca area remained significant after adjusting antipsychotic medication doses, suggesting that the medication’s effects on hemodynamics in specific language-related brain regions is independent of structural changes [33].

fNIRS Evaluation of Psychotherapy or Neurofeedback Treatment

fNIRS provides a real-time, dynamic tool for monitoring the efficacy of psychotherapy and neurofeedback (NFB) treatment, particularly in modulating prefrontal and temporal functions. fNIRS-NFB, by targeting the regulation of superior temporal gyrus (STG) and anterior cingulate activity, improves treatment-resistant auditory verbal hallucinations (AVH). A case study by Storchak et al showed that downregulating STG activity guided by fNIRS reduced the frequency of hallucinations [34]. Hirano et al further pointed out that while fMRI-NFB offers more precise control of local activity, fNIRS is more promising in terms of portability and real-time performance [35]. Balconi et al, combining fNIRS with EEG, found that after emotional NFB training, patients’ prefrontal activity became more balanced, and improvements in HbO fluctuations preceded changes in overt behavioral scores, highlighting the sensitivity of fNIRS to early treatment responses [36]. Bhargav et al found that bilateral prefrontal activation in schizophrenia patients during yoga breathing was insufficient, while HbO in healthy controls increased significantly, suggesting that such interventions may need to be combined with medication to optimize prefrontal hemodynamic responses [32]. Kumar et al proposed that delayed compensatory activation in the right frontal pole during working memory tasks may be a potential target for cognitive training [22].

The use of fNIRS in treatment monitoring highlights several research directions. The heterogeneity of drug effects requires longitudinal studies to compare the differential effects of different antipsychotic drugs on prefrontal and temporal hemodynamic patterns. Individualized intervention with FNIR-based NFB can precisely regulate symptom-specific brain regions, and fNIRS-guided NFB can reduce hallucinations in AVH patients by 60% (such as STG for AVH and DLPFC for negative symptoms) but requires an optimized task paradigm (such as combined n-back and VFT). Multimodal integration combining EEG/fMRI can complement spatial and temporal resolutions, as Balconi et al revealed neural dynamics in emotion processing through synchronized fNIRS-EEG monitoring [36]. Future research should expand sample sizes, explore fNIRS biomarkers (eg, oxy-Hb change rates) as predictors of treatment efficacy, and develop portable fNIRS devices to facilitate clinical translation. The information of all included studies is summarized in Table 2.

Table 2.

Summary of fNIRS studies in schizophrenia research.

Author	Year	Research method	Key findings	Conclusion
Kumar V, et al	2021	fNIRS combined with N-back task	Patients with schizophrenia showed lower deoxyhemoglobin concentration in the right frontopolar cortex (BA10), and olanzapine dosage correlated negatively with right frontopolar cortex activation	Delayed but compensatory hyperactivation in the right frontopolar cortex may underlie working memory deficits in schizophrenia
Ma CC, et al	2024	fNIRS combined with 2-back task	The 2-back task significantly activated the left DLPFC	The 2-back task can be combined with tDCS for treating negative symptoms in schizophrenia
Hirata K, et al	2018	fNIRS combined with social and non-social cognitive tasks	Patients with schizophrenia showed significantly reduced activation in the left frontotemporal area compared to healthy controls	Schizophrenia and autism spectrum disorder have distinct brain pathophysiologies in cognitive processing
Chou PH, et al	2020	fNIRS combined with clinical assessment	fNIRS can serve as a potential biomarker for schizophrenia	fNIRS has broad prospects in psychiatry and requires further optimization
Wei Y, et al	2022	fNIRS combined with VFT	Individuals at clinical high risk for psychosis showed significantly reduced activation in the rSTG	Abnormal activation in the rSTG may be associated with working memory deficits in the early stages of psychosis
Singh R, et al	2025	fNIRS combined with EEG and facial features	Multimodal data (fNIRS and EEG) improved the classification of first-episode psychosis	Multimodal approaches are advantageous for predicting early psychosis
Gao C, et al	2025	fNIRS combined with TBS treatment	TBS improved negative symptoms in chronic schizophrenia, with fNIRS showing changes in brain activity	TBS is an effective method for improving negative symptoms in chronic schizophrenia
Zouraraki C, et al	2023	fNIRS combined with resting-state functional imaging	Individuals with high schizotypal traits and schizotypal personality disorder showed functional alterations in the striatum, frontal, and temporal regions	Brain dysfunctions are evident in the subclinical part of the schizophrenia spectrum
Cattarinussi G, et al	2024	fNIRS combined with multiple neuroimaging techniques	Catatonia is characterized by structural, functional, perfusion, and metabolic cortico-subcortical abnormalities	Further research is needed to clarify the neural mechanisms underlying catatonia
Akın A	2021	fNIRS combined with Stroop task	A new NCR was proposed, showing high sensitivity and specificity	NCR can be a reliable biomarker for neuropsychiatric diseases
Storchak H, et al	2019	fNIRS combined with neurofeedback	fNIRS neurofeedback reduced auditory verbal hallucinations	fNIRS neurofeedback has potential in treating hallucinations
Curtin A, et al	2019	fNIRS monitoring of bilateral prefrontal cortex activity combined with vSAT	Activity in the left MFG correlated strongly with PANSS negative symptom scores in patients with schizophrenia	Abnormal activation in the left MFG may indicate attention network dysregulation linked to negative symptoms
Bhargav H, et al	2014	fNIRS monitoring of frontal hemodynamic responses to high-frequency yoga breathing	Patients with schizophrenia showed significantly lower bilateral prefrontal activation during high-frequency yoga breathing compared to healthy controls	Hypo-frontality in schizophrenia patients during high-frequency yoga breathing may support future diagnosis
Iwashiro N, et al	2016	fNIRS and MRI combined with VFT	UHR and FES groups showed significantly reduced brain activity in the left PT, and activity correlated positively with volume in the left PT	The relationship between neural activity and gray matter volume in the left PT may reflect a specific pathophysiology related to schizophrenia onset
Kopf J, et al	2011	fNIRS combined with N-back task	Short allele carriers showed greater increases in oxyhemoglobin in task-related areas, possibly indicating a compensatory mechanism	NOS1 gene polymorphism may influence prefrontal oxygenation during working memory tasks
Sayar-Akaslan D, et al	2021	fNIRS combined with theory of mind task	Cortical activity in patients with schizophrenia and bipolar disorder differed from healthy controls	Schizophrenia and bipolar disorder may share common neurobiological mechanisms in social cognition processing
Hirano Y and Tamura S	2021	fNIRS and EEG combined with neurofeedback	fMRI-based NF can alleviate treatment-resistant AVH, while EEG-based neurofeedback failed to modulate auditory-evoked potentials	Real-time neurofeedback training can help reduce severe symptoms and improve social functioning
Balconi M, et al	2018	fNIRS and EEG combined with emotion regulation task	Implicit measures showed modulation and improved capabilities after NF training, while explicit measures correctly identified emotional valence before and after treatment	Emotion processing showed abnormalities in spatial and temporal expression, but explicit evaluation of emotional stimuli was preserved

Table 2 summarizes 18 functional near-infrared spectroscopy (fNIRS) studies investigating schizophrenia, including research methodologies, key neuroimaging findings, and clinical conclusions. Studies are ordered chronologically (2011–2025) and highlight fNIRS applications in: working memory assessment (N-back tasks), social cognition, neurofeedback interventions, multimodal integration (EEG/MRI), biomarker development, and treatment monitoring (TBS/tDCS/yoga). Key patterns include prefrontal hypo-frontality, compensatory activation mechanisms, and distinct functional alterations across schizophrenia spectrum disorders. fNIRS – functional near-infrared spectroscopy; BA – Brodmann area; DLPFC – left dorsolateral prefrontal cortex; tDCS – transcranial direct current stimulation; TBS – theta burst stimulation; VFT – verbal fluency test; EEG – electroencephalography; MRI – magnetic resonance imaging; NCR – neurocognitive ratio; PANSS – Positive and Negative Syndrome Scale; SANS – Scale for the Assessment of Negative Symptoms; UHR – ultra-high risk; FES – first-episode schizophrenia; PT – pars triangularis; MFG – middle frontal gyrus; vSAT – visual spatial attention task; AVH – auditory verbal hallucinations; NF – neurofeedback; NOS1 – nitric oxide synthase 1.

Methodological Challenges and Controversies

Technical Limitations

Signal Noise Processing (eg, Head Motion, Physiological Artifacts)

The application of fNIRS in schizophrenia research is limited by signal noise interference. Physiological artifacts (eg, respiration, heartbeat) and signal drift caused by head motion are major challenges. For example, Li et al found that in the VFT of schizophrenia patients, the frequency of head motion was significantly higher than in healthy controls, leading to increased instability in prefrontal oxy-Hb signals. Although sliding window averaging and bandpass filtering (eg, 0.01–0.1 Hz) can partially mitigate noise, it is still difficult to completely eliminate interference in complex motor tasks (eg, emotional expression experiments) [37]. Additionally, Akın proposed the “neurocognitive ratio” as a standardized metric, but the effectiveness of this method is highly dependent on the robustness of noise suppression algorithms, which have limited adaptability to individual differences (eg, scalp thickness and hair density) [20].

Insufficient Ability to Detect Deep Brain Regions

The light penetration depth of fNIRS is typically limited to the cortical surface (about 2–3 cm), making it difficult to detect activity in deep nuclei (eg, thalamus, amygdala). In a study of first-episode schizophrenia in adolescents, Zhang et al found that abnormal prefrontal activation was significant during a verbal fluency task, but it was impossible to assess limbic system functions closely related to symptoms, limiting the completeness of mechanistic explanations [38]. Although multi-channel arrays (eg, 48 channels) can partially improve spatial resolution [39], deep-signal attenuation still makes functional connectivity analysis of key brain regions impossible.

Research Heterogeneity

Differences in Experimental Design

Significant differences in task paradigms and control group designs across studies reduce result comparability. For example, Yeung’s [40] systematic review pointed out that differences in vocabulary types (eg, semantic categorization vs letter generation) and duration settings (30 seconds to 2 minutes) in the VFT significantly affect prefrontal activation patterns, with some studies even reporting opposing conclusions (eg, enhanced or reduced activation in patients with negative symptoms). Wang et al used novel emotional expression stimuli (eg, family conflict scenarios), while traditional studies mostly employed standardized emotional images (eg, IAPS), and differences in task ecological validity may confuse neural activation characteristics [37]. Wei et al found that if the education level and cognitive abilities of healthy control groups were not strictly matched, it might exaggerate the “low prefrontal activation” effect in patient groups [41].

Standardization Issues in Data Analysis Methods

fNIRS data analysis lacks unified standards, particularly in signal preprocessing and statistical methods. Some studies use general linear models (GLM) [42] for denoising, while others rely on independent component analysis (ICA) [39], making cross-study results difficult to integrate. In terms of statistical thresholds, the choice of multiple comparison correction methods significantly affects significance determination. For example, Zhang et al found no difference in temporal lobe activation after FDR correction, while Bonferroni correction may be overly conservative [38].

Biomarker Exploration

Existing studies preliminarily support the potential of fNIRS as an auxiliary diagnostic tool but still face the following challenges. In diagnostics, Erdoğan et al classified 4 types of neuropsychiatric diseases using machine learning, and found a sensitivity of 78% for prefrontal oxy-Hb time series, but the accuracy of distinguishing between schizophrenia and bipolar disorder was only 65%, suggesting the need for combining multimodal indicators (eg, EEG) [42]. In prognostic prediction, Wei et al found that hypoactivation in the rSTG of CHR individuals could predict conversion to schizophrenia (AUC=0.72), but longitudinal validation samples were insufficient [41]. In terms of treatment response biomarkers, Akın [20] proposed that the “neurocognitive ratio” (correlation between prefrontal activation and task performance) might predict tDCS efficacy, but the stability of this metric across tasks (eg, n-back vs VFT) has not been verified.

Despite the significant advantages of fNIRS in portability and ecological validity, its technical limitations and research heterogeneity restrict the clinical translation of biomarkers. Future efforts should focus on developing adaptive noise suppression algorithms (eg, combining motion tracking and synchronized physiological signal recording), establishing international consensus on task paradigms and data analysis (eg, COBIDAS-fNIRS), exploring indirect inference methods for deep brain regions (eg, combining DTI fiber tracking), and conducting large-sample multicenter studies to validate the generalizability of biomarkers (eg, ENIGMA-Schizophrenia Consortium).

Future Research Directions

Technological Innovation

High-Density fNIRS and Multimodal Integration

Previous studies have preliminarily verified the potential of high-density fNIRS (eg, 52-channel systems) in localizing symptom-related brain regions (eg, the right postcentral gyrus) [43]. Future integration with EEG or fMRI can complement spatial and temporal resolutions, such as synchronized EEG-fNIRS recording to simultaneously capture neural electrical activity and hemodynamic responses, optimizing monitoring of emotional regulation or cognitive interventions [44]. Multimodal integration (eg, fNIRS with DTI) can explore abnormalities in white-matter fiber bundles and functional connectivity [45].

Application of Machine Learning in Data Analysis

Machine learning has shown high precision in fNIRS signal classification [46]. Deep neural networks (DNNs) can distinguish people experiencing first-episode schizophrenia from healthy individuals (AUC=0.89), with feature importance analysis indicating that dynamic patterns of prefrontal oxy-Hb are key [47]. Dynamic functional connectivity (DFC) combined with support vector machines (SVMs) can improve classification accuracy between schizophrenia and bipolar disorder to 82% [48].

Clinical Translation

Individualized treatment strategies based on fNIRS: For treatment-resistant auditory hallucinations (TRAVH), hypoactivation in the right postcentral gyrus (ch22) can serve as a target for tDCS or rTMS [43]. Medication response prediction: The correlation between prefrontal activation and antipsychotic dosage (eg, negative correlation between olanzapine equivalent dose and deoxy-Hb in the right frontal pole) may guide dosage adjustment [22]. Early screening in high-risk populations: Hypoactivation in the rSTG of CHR individuals (AUC=0.72) can be combined with genetic risk scores (eg, COMT gene polymorphisms) to enhance prediction specificity [45]. Prefrontal activation patterns in the VFT (eg, Δβ values) may serve as conversion biomarkers [42]. Despite promising classification accuracy, fNIRS biomarker generalizability is limited by small sizes and cultural/language biases in VFT paradigms. Multicenter standardization (eg, ENIGMA-Schizophrenia) is urgently needed [28].

Interdisciplinary Collaboration

Combining genetics and behavior: Through multi-omics integration, fNIRS biomarkers (eg, DLPFC activation) combined with genome-wide association studies (GWAS) can parse gene-brain-behavior pathways (eg, the impact of DISC1 mutations on prefrontal plasticity). Quantification of behavioral phenotypes: Virtual reality (VR) tasks simulating social scenarios, with synchronized recording of fNIRS and eye movement data, can reveal the behavioral-neural mechanisms of negative symptoms. Multidimensional data modeling: Using fNIRS time series to construct brain network dynamics models to predict symptom fluctuations (eg, periodic worsening of negative symptoms) [44]. Ecological momentary assessment (EMA): Combining smartphone apps to collect patients’ emotional states and correlate with fNIRS data (eg, prefrontal activation changes induced by stress events).

Declaration of Figures’ Authenticity

All figures submitted have been created by the authors who confirm that the images are original with no duplication and have not been previously published in whole or in part.