Introduction

Chronic insomnia disorder(CID) is a prevalent condition that significantly impacts patients’ quality of life and cognitive function1. It is characterized by difficulties initiating or maintaining sleep, leading to daytime dysfunction and cognitive impairment2. Approximately 10% of the adult population suffers from an insomnia disorder, and another 20% experiences occasional insomnia symptoms. Insomnia is often a chronic condition, with a 40% persistence rate over 5 years2. It is a significant public health problem that should be addressed at the individual level with appropriate clinical care and at the population level with large-scale sleep health interventions3. Insomnia may play an important role as a risk factor, a comorbid condition, and a transdiagnostic symptom for many mental disorders, including mood/anxiety disorders and schizophrenia4.

The neural basis of cognitive impairments in CID involves dysregulation in prefrontal cortex (PFC) activation and functional connectivity(FC)5,6,7. The PFC plays a critical role in executive functions, including verbal fluency, which is often compromised in individuals with insomnia8. Previous studies have shown reduced activation in the dorsolateral prefrontal cortex (DLPFC) and medial prefrontal cortex (mPFC) in insomnia patients during cognitive tasks9. Cognitive dysfunction in chronic insomnia patients manifests as deficits in working memory, executive function, and attention10. Neuroimaging studies have implicated prefrontal cortex dysfunction in the pathophysiology of chronic insomnia, highlighting its role in cognitive deficits11,12. Additionally, altered within and between hemispheres has been observed, suggesting a broader network dysfunction13. These findings highlight the importance of investigating both activation patterns and connectivity profiles to fully understand the neural mechanisms contributing to cognitive deficits in CID.

Functional near-infrared spectroscopy (fNIRS) has emerged as a valuable tool for investigating cortical activation and FC in real-time during cognitive tasks14,15. It offers a non-invasive and cost-effective means of assessing neurohemodynamic changes in the prefrontal cortex, which is central to executive function16. Recent studies using fNIRS have reported reduced prefrontal activation in chronic insomnia patients during verbal fluency tasks (VFTs), suggesting a potential neurobiological basis for their cognitive deficits7,9,17. Furthermore, fNIRS enables the assessment of FC by analyzing correlations in hemodynamic responses between different brain regions8. While previous fNIRS studies have established the presence of prefrontal dysfunction during cognitive tasks in insomnia, they have typically focused on either cortical activation or functional connectivity in isolation, often in smaller samples. The present study aims to extend this work by utilizing multi-channel fNIRS to simultaneously investigate both cortical activation and functional connectivity patterns during a VFT in a relatively large cohort of CID patients and healthy controls. Furthermore, we seek to explicitly link these neural measures to specific clinical sleep profiles and evaluate their potential diagnostic discriminative capacity. This integrated approach is designed to provide a more comprehensive pathophysiological model of cognitive impairment in CID.

Methods

Patients and experimental protocol

Inpatients diagnosed with CID who were treated in the Sleep Department of Zigong Mental Health Center from June 2024 to February 2025 were selected. Inclusion criteria: (1) Meeting the Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-5) criteria for CID; (2) The Pittsburgh Sleep Quality Index (PSQI) score was more than 7 points. (3) years of education ≥ 6 years; (4) voluntary participation in this study and signed informed consent. (5) right-handed; (6) able to complete the experimental tasks. Exclusion criteria: (1) combined with physical diseases leading to sleep disorders or other mental disorders or substance abuse; (2) Hamilton Depression Scale-17 item (HAMD-17) score ≥ 17;3. Hamilton Anxiety Scale (HAMA) score ≥ 14; All participants underwent assessment using the PSQI to evaluate sleep quality18. The experimental protocol was approved by the Ethics Committee of Zigong Hospital Affiliated to Southwest Medical University. During the experiment, participants were seated comfortably in a quiet room and performed a VFT19,20 while their cortical activation was measured using fNIRS. The VFT consisted of a 30-second pre-task baseline, a 60-second task period, and a 60-second post-task baseline. In a baseline session before the task, participants were asked to count out loud to 16; However, at the baseline phase after the task, they were asked to count to 32. Throughout the task, participants were instructed to use four Chinese characters to generate as many Chinese phrases as possible, such as “tian” (“ day “), “bai” (“ white “), “shu” (“ tree “), and “hua” (“ flower “). The presentation order of these four characters was counterbalanced across all participants using a Latin square design to control for potential effects of character-specific properties (e.g., lexical frequency) on task performance and cortical activation. During this period, automatic changes were made every 15 s. All patients received detailed instructions on the experimental procedures before the start of the study.

NIRS measurement

Cortical activation was measured using a 53-channel fNIRS device (BS-3000, Wuhan Znion Medical Technology Co., Wuhan, China) with a sampling rate of 20 Hz. The near-infrared light wavelengths used were 690 and 830 nm. Channels were placed according to the 10–20 system, with source-detector pairs positioned 3 cm apart(source 9 is in the FPz). Prior to each measurement, the fNIRS system underwent calibration, and testing would only commence when a repetition rate of over 90% was achieved. The channel distribution can be found in Supplementary Material 1.

Data analysis

Data pre-processing

Firstly, the data sampling rate was reduced to 10HZ. 10s before the task was used as the baseline, and the Coefficient of Variation (CV) value of the data was calculated. If the CV value of a channel was greater than 25%, it was considered a bad channel. The CV values of each task period were calculated, and if the CV value of a channel was greater than 25%, it was considered a bad channel21. Using the mark as the starting point, 2 s before and 60 s after each mark were taken for overlay averaging, and 2 s before the mark was baseline calibrated for the data of this block. Raw optical intensity data were first converted to optical density using a MATLAB-based toolkit, Homer222. tMotion was assigned to 0.3 s and tMask to 3 s, with the remaining motion correction parameters kept at default. To detect and fix motion artifacts, the process involved applying a moving standard deviation combined with cubic spline interpolation. Additionally, a bandpass filter within the range of 0.01 to 0.1 Hz was utilized to eliminate physiological noise, which encompasses low-frequency drift and high-frequency noise. Changes in oxyhemoglobin (Oxy-Hb), deoxyhemoglobin (Deoxy-Hb), and total hemoglobin (total-Hb) concentrations were computed according to the modified Beer-Lambert Law23.

General linear model (GLM) analysis

Task-evoked activation during the VFT was examined using a GLM24,25 approach with a MATLAB-based toolbox, NIRS-KIT26. The experimental design matrix included a constant term and a task regressor constructed by convolving a boxcar function with the canonical hemodynamic response function (HRF)27. Task β-values reflecting task-evoked activation were derived and used in group-level statistical analyses26,28. Because Oxy-Hb has a typically superior signal-to-noise ratio and a stronger correlation with cerebral blood flow than Deoxy-Hb, Oxy-Hb was selected as the primary measure for our subsequent analyses29,30. Given the reported lack of satisfactory stability in single-channel level signals from previous studies, we conducted ROI(regions of interest)-level analysis. The integrated value over a 60-second task period was computed for each channel and averaged within a given ROI to serve as the ROI indicator.

ROI-level feature extraction

The integrated value of Oxy-Hb concentration changes over the 60-second task period was computed for each channel. Six ROIs were defined a priori based on the Automated Anatomical Labeling (AAL) atlas. The assignment of channels to these ROIs was performed probabilistically by mapping the pre-defined Montreal Neurological Institute (MNI) coordinates of each channel’s midpoint (provided in Supplementary Material 2, Table 1.1) onto the AAL parcellation. The average value of all channels within a given ROI was calculated as the ROI metric, and any CV value exceeding 25% was considered a “bad channel” and would be omitted from the ROI calculation. Six rois were identified based on the Automated Anatomical Labeling (AAL) probabilistic map: left inferior frontal gyrus (IFG_L), left middle frontal gyrus (MFG_L), left superior frontal gyrus (SFG_L), right superior frontal gyrus (SFG_R), right middle frontal gyrus (MFG_R), and right inferior frontal gyrus (IFG_R). The Oxy-Hb concentration changes in these ROIs were analyzed to assess task-related activation. See Supplementary Material 2 for the distribution of ROI.

Statistical analysis

Statistical analysis was performed using IBM SPSS Statistics (Version 26). Group differences in age and educational level were assessed using the two-sample t-test, while gender differences were quantified using the chi-square test. One-sample t-tests were conducted on task β-values to assess group-level task-evoked activation within each group and channel. Group differences in activation were calculated by a two-sample T-test. For the integrated values at the ROI level, two-sample t-tests were conducted for each ROI. Statistical tests were adjusted for multiple comparisons with the use of the false discovery rate (FDR). The Pearson correlation coefficient of PSQI total score, factor score of CID, and HBO activation in the ROI brain region was calculated to explore the relationship between VFT activation and insomnia. This correlation analysis was performed only for ROIs with significant between-group differences. Given the exploratory nature of these correlation analyses, p-values are reported without correction for multiple comparisons, and the results should be interpreted with caution as generating hypotheses for future research. All were two-tailed tests, and p < 0.05 was considered statistically significant.

Multivariate and combined model analysis

To further evaluate the discriminative power of fNIRS metrics and their integration with behavioral and clinical data, we performed supplementary classification analyses. First, a multivariate logistic regression model was constructed using the Oxy-Hb activation values from all six ROIs as simultaneous predictors. Second, a combined model was developed by integrating the multivariate fNIRS data with behavioral performance (the total number of correct words generated during the VFT) and clinical scale scores (HAMA and HAMD). The performance of these models in distinguishing CID patients from HC was assessed using ROC analysis, and their AUC values were compared to the single-ROI models.

Result

Demographic and PSQI scores

The study recruited 81 patients with CID and 100 HC. The demographics and clinical characteristics of the participants are summarized in Table 1. The two groups were comparable in terms of gender distribution, age, and education level. Specifically, there were no significant differences in the proportion of males and females between the chronic insomnia group and the HC group (p = 0.166). Similarly, the mean age and years of education did not differ significantly between the two groups (p = 0.080 and p = 0.177, respectively).

Table 1 Demographic characteristics and clinical scale assessment.
Full size table

Brain activation patterns during VFT

Figure 1 presents the brain activation patterns in HC and patients with CID during a VFT, as assessed by fNIRS. The topographical and 3D brain model visualizations reveal distinct differences in cortical activation between the two groups. In the HC group, robust activation was observed across multiple channels, particularly in the prefrontal and temporal regions, with extensive areas showing high activation levels. Specific channels showing significant activation (after FDR correction) include channels 3 (P < 0.01,T = 2.816), 7 (P < 0.01, T = 7.742), 8(P = 0.02,T = 2.289),13 (P < 0.01, T = 3.358), 16 (P < 0.01, T = 5.9473), 22 (P < 0.01, T = 4.5087), 23 (P < 0.01, T = 6.8839), 24 (P < 0.01, T = 2.706), 25 (P = 0.01, T = 2.531), 27 (P < 0.01,T = 7.351), 28 (P < 0.01, T = 5.076), 29 (P < 0.01,T = 5.326), 30 (P < 0.01,T = 6.497), 31 (P < 0.01,T = 3.258),33 (P < 0.01,T = 2.804),34 (P < 0.01, T = 4.658), 35 (P < 0.01, T = 8.113), 36 (P < 0.01,T = 4.100), 37 (P < 0.01,T = 9.8074), 41 (P = 0.02, T = 2.438),42 (P < 0.01, T = 4.3144), 43 (P < 0.01, T = 4.748), 44 (P < 0.01, T = 3.904), 46 (P < 0.01,T = 3.3799), 48 (P < 0.01,T = 6.343), 49 (P < 0.01,T = 2.6981), and 50 (P < 0.01,T = 5.7313).In contrast, the CID group exhibited significantly reduced activation across most channels, with fewer areas showing high activation levels and larger regions demonstrating low activation (primarily blue colors). Only a few channels showed significant activation after FDR correction, including channel 23 (P < 0.01, T = 3.7841).

Fig. 1
Fig. 1
Full size image

Cortical Activation Patterns During VFT in HC and CID Groups (FDR-corrected, p < 0.05). This figure depicts the spatial distribution of significant cortical activation (oxy-hemoglobin changes) during the VFT in HC(top panels) and patients with CID (bottom panels), corrected for multiple comparisons using the FDR method (p < 0.05). The left (L) and right (R) hemispheres are shown separately. In the HC group, robust activation is observed across prefrontal and temporal regions, with widespread high-intensity signals (red-yellow hues). In contrast, the CID group exhibits markedly reduced activation, limited to sparse clusters (blue-green hues), particularly in the left mPFC and bilateral DLPFC. The color bar indicates activation intensity (ΔHbO₂), with warmer colors representing higher activation.

Differences in brain activation between HC and CID during VFT

Figure 2 illustrates the differences in brain activation between HC and patients with CID during a VFT, as measured by fNIRS. The topographical map shows that the HC group exhibited significantly higher activation in multiple channels compared to the CID group, particularly in the prefrontal and temporal regions. Significant differences were detected in channels 7 (P < 0.01, T = 4.098), 13 (P < 0.01, T = 3.236), 16 (P < 0.01, T = 4.194), 16 (P = 0.04, T = 2.069),22 (P < 0.01, T = 3.665), 23 (P < 0.01, T = 3.9915), 27 (P < 0.01, T = 2.772), 29 (P < 0.01, T = 3.814), 30 (P < 0.01, T = 3.612), 33 (P = 0.01, T = 2.523), 35 (P < 0.01, T = 4.3505), 36 (P < 0.01, T = 3.559), 37 (P < 0.01, T = 6.8111), 42 (P < 0.01, T = 2.672), 43 (P < 0.01, T = 2.799), 44 (P < 0.01, T = 3.789), 46 (5 P < 0.01, T = 2.958), and 48 (P < 0.01, T = 4.314). The 3D brain model further emphasizes the prominent activation differences in the left mPFC and bilateral DLPFC, as well as the left temporal lobe. These findings indicate that patients with chronic insomnia have weaker cortical activation compared to HC during the VFT.

Fig. 2
Fig. 2
Full size image

Group Differences in Cortical Activation Between HC and CID During VFT (FDR-corrected, p < 0.05). The left panel displays a topographical representation of the activation patterns, with warmer colors (red and yellow) indicating higher activation levels and cooler colors (blue and green) indicating lower activation levels. The right panel presents a 3D brain model, highlighting the activation areas in the left (L) and right (R) hemispheres. The topographical map shows robust activation across multiple channels, particularly in the prefrontal and temporal regions. Significant activation clusters are evident in the left mPFC and bilateral DLPFC, as well as the left temporal lobe. The 3D brain model further emphasizes the prominent activation in these regions, with high activation levels represented by red and yellow areas.

Regional differences in brain activation between HC and CID

Figure 3. Regional Differences in Oxy-Hemoglobin (HbO₂) Levels Between HC and CID Groups Across Six ROIs. The figure uses violin plots to illustrate the distribution of oxy-hemoglobin (ROI_HBO) levels across six regions of interest (ROIs): IFG_L, MFG_L, SFG_L, SFG_R, MFG_R, and IFG_R. Each violin plot shows the data distribution for both HC and CID groups, with red and yellow colors representing HC and CID, respectively. The results indicate significant differences in activation between the two groups across all six ROIs (all p < 0.01). In each ROI, the HC group consistently showed higher activation levels compared to the CID group.

Fig. 3
Fig. 3
Full size image

illustrates the regional differences in brain activation between HC and patients with CID during a VFT, focusing on six regions of interest (ROIs). The violin plots show the distribution of oxy-hemoglobin levels for each group. The results showed significant differences in activation between the HC and CID groups in all six ROIs, p < 0.01. In each ROI, the HC group consistently demonstrated higher activation levels compared to the CID group.

Correlation between brain activation and sleep parameters

Figure 4 illustrates the correlation between brain activation and sleep parameters in patients with CID during a VFT. The left panel shows a negative correlation between activation in the IFG_L and sleep latency (R=-0.241, Puncorrected<0.05), indicating that higher activation in IFG_L is associated with shorter sleep latency. Additionally, a positive correlation was observed between IFG_L activation and sleep duration (R = 0.329, Puncorrected<0.01), suggesting that increased activation in this region is linked to longer sleep duration. The right panel demonstrates positive correlations between activation in the SFG_R and both the PSQI total score (R = 0.231, Puncorrected<0.05) and sleep efficiency (R = 0.357, Puncorrected<0.01). These findings indicate that brain activation during the VFT is associated with specific sleep parameters in CID.

Fig. 4
Fig. 4
Full size image

Correlations between cortical activation and sleep parameters in CID. This figure displays the relationships between oxy-hemoglobin (HbO₂) activation in specific brain regions and clinical sleep metrics in CID. Left panel: Activation in the IFG_L negatively correlates with sleep latency (R = -0.241, P < 0.05) and positively correlates with sleep duration (R = 0.329, *P < 0.01), suggesting that higher IFG_L activation is associated with shorter time to fall asleep and longer total sleep time. Right panel: Activation in the SFG_R positively correlates with both the PSQI total score (R = 0.231, P < 0.05) and sleep efficiency (R = 0.357, *P < 0.01), indicating that increased SFG_R activation relates to poorer subjective sleep quality but improved sleep maintenance.

FC differences between HC and CID

Figure 5 presents the FC matrices for HC and CID during a VFT, as assessed by fNIRS. The top two panels show the FC of the HC and CID groups, respectively. The FC matrix between the ROIs is shown in the lower left panel. The right panel shows the FC matrix for all 53 channels. These matrices have been corrected using the FDR method to account for multiple comparisons and highlight significant differences between the two groups as determined by two independent sample t-tests. There were strong connections between multiple rois, especially between IFG_L and MFG_L(T = 3.180,P = 0.002), IFG_L and SFG_R(T = 3.869,P < 0.001), MFG_L and SFG_R(T = 4.095,P < 0.001). After FDR calibration, the differences were statistically significant. The right panel shows that the two group differences show differences between multiple channels, passing FDR calibration, and that CID patients have weaker FC than HC during VFT.

Fig. 5
Fig. 5
Full size image

FC differences between HC and CID. This figure presents the FC matrices for HC and patients with CID during a VFT. The matrices illustrate the connectivity strength between different brain regions, with warmer colors (red and yellow) indicating stronger connectivity and cooler colors (blue and green) indicating weaker connectivity. The left panel shows the FC matrix between six regions of interest (ROIs): (1) IFG_L, (2) MFG_L, (3) SFG_L, (4) SFG_R, (5) MFG_R, (6) IFG_R. The right panel displays the FC matrix across all 53 channels. These matrices have been corrected using the FDR method to account for multiple comparisons and highlight significant differences between the two groups as determined by two independent sample t-tests.

Diagnostic accuracy of brain activation in chronic insomnia

Figure 6 presents receiver operating characteristic (ROC) curves for different ROIs in distinguishing patients with CID from HC based on brain activation during a VFT. Each curve represents the relationship between sensitivity and 100%-specificity for each ROI, with the area under the curve (AUC) indicating the diagnostic performance. The AUC values for each ROI are as follows: IFG_L (AUC = 0.6807, P < 0.01), MFG_L (AUC = 0.6981, P < 0.01), SFG_L (AUC = 0.7936, P < 0.01), SFG_R (AUC = 0.7759, P < 0.01), MFG_R (AUC = 0.7027, P < 0.01), and IFG_R (AUC = 0.6953, P < 0.01). These results indicate that activation levels in these frontal regions during the VFT can serve as potential biomarkers for distinguishing CID patients from HC individuals. To provide a more comprehensive assessment of classification performance, we calculated additional metrics including accuracy, precision, recall (sensitivity), specificity, and F1-score at the optimal cut-off point (Supplementary Table 1). The left superior frontal gyrus (SFG_L) demonstrated the most balanced performance across all metrics, with an AUC of 0.7936, accuracy of 73.5%, sensitivity of 77.8%, and specificity of 70.0%. These results indicate that activation levels in these frontal regions during the VFT show promising, though not perfect, capacity for distinguishing CID patients from HC individuals.

Fig. 6
Fig. 6
Full size image

ROC Curves for ROIs in Distinguishing CID from HC. This figure illustrates the diagnostic performance of six frontal ROIs in differentiating CID from HC)based on cortical activation during the VFT. ROC curves are plotted for each ROI, with sensitivity (%) on the y-axis and 100% - specificity (%) on the x-axis. The area under the curve (AUC) values and statistical significance are as follows: IFG_L, AUC = 0.681, P < 0.01, MFG_L, AUC = 0.698, P < 0.01, SFG_L, AUC = 0.7936, P < 0.01, SFG_R, AUC = 0.7759, P < 0.01, MFG_R, AUC = 0.7027, P < 0.01, and IFG_R, AUC = 0.6953, P < 0.01. The dashed diagonal line represents chance-level discrimination (AUC = 0.5).

Enhanced classification using multivariate and combined models

To explore whether integrating multiple neural and non-neural features could improve group discrimination, we developed a series of advanced classification models. A multivariate model incorporating activation from all six prefrontal ROIs achieved a higher AUC of 0.842 (95% CI: 0.783–0.901) compared to the best single-ROI model (SFG_L, AUC = 0.794). Furthermore, a combined model that additionally included verbal fluency performance and scores from anxiety and depression scales demonstrated excellent discriminative capacity, with an AUC of 0.912 (95% CI: 0.871–0.953). Detailed performance metrics for all models are presented in Supplementary Table 2.

Discussion

The present study utilized multi-channel fNIRS to investigate cortical activation and FC patterns during a VFT in patients with CID, revealing distinct neural alterations compared to HC. Beyond replicating findings of prefrontal hypoactivation, our study provides several novel insights: first, by demonstrating concurrent reductions in both local prefrontal activation and global functional connectivity, we underscore the multi-level nature of neural disruption in CID; second, we establish clinically meaningful correlations between specific neural deficits and quantifiable sleep parameters; and third, we quantitatively evaluate the potential of these fNIRS-derived metrics as diagnostic biomarkers. Collectively, these findings offer a more integrative neuropathological framework for understanding CID. Key findings demonstrated significantly reduced prefrontal and temporal cortical activation in CID patients, particularly in the left mPFC, bilateral DLPFC, and left temporal regions, alongside weakened FC between critical frontal regions such as the inferior frontal gyrus (IFG), middle frontal gyrus (MFG), and superior frontal gyrus (SFG). Notably, these activation deficits correlated with clinical sleep parameters, including prolonged sleep latency and reduced sleep efficiency, while ROC analysis identified frontal activation patterns as potential diagnostic biomarkers for CID. Collectively, these results underscore impaired prefrontal executive function and disrupted functional network integration as neural hallmarks of chronic insomnia, providing novel insights into its neuropathological mechanisms.

Our study aligns with previous research in showing reduced prefrontal cortex (PFC) activation in patients with chronic insomnia during cognitive tasks8. For instance, Zhou et al. and Herrmann et al. also observed reduced activation in the DLPFC and mPFC during similar tasks9,31. This aligns with the findings of Noda et al.32, who reported similar activation patterns in patients with major depressive disorder (MDD) and insomnia. Our results further highlight the potential of fNIRS as a tool for assessing neurophysiological changes in insomnia, offering insights for future research and clinical applications7,15,32,33. Similar to the present study, other studies7 have found that oxyhemoglobin (oxy-Hb) activation in the left inferior frontal gyrus (IFG_L) is significantly reduced in CID patients, and this decrease is negatively correlated with sleep latency. (r=-0.41, P < 0.05). The activation intensity of IFG_L was positively correlated with total sleep time (TST) (r = 0.38, P < 0.05). This suggests that the functional inhibition of IFG_L may contribute to the difficulty in falling asleep and sleep maintenance impairment through impaired attention and verbal execution. SFG_R activation was restored and positively correlated with sleep efficiency improvement (r = 0.52, P < 0.01). Similar to our study, Gong HY’s study also found that SFG_R activation was positively correlated with sleep efficiency (SE) improvement (r = 0.52, P < 0.01)15. However, in the fMRI study by Zhang et al., SFG_R activation was negatively correlated with sleep efficiency. This difference may be related to the type of experimental task and the characteristics of the population34. It is important to note that these correlational analyses between brain activation and sleep parameters were exploratory in nature and were not adjusted for multiple comparisons. Therefore, these particular findings should be considered preliminary and require replication in future confirmatory studies. In a magnetic resonance study35, it was found that the FC strength of the right DLPFC in CID patients was significantly reduced (P < 0.05), and the area under the ROC curve (AUC) was 0.793, indicating that the functional connectivity strength of the right DLPFC reached a “good” level in distinguishing CID patients from healthy controls. Similarly, cortical activation of the left and right SFG, in particular, showed good discrimination between groups (AUC = 0.7936). When evaluated with additional metrics, SFG_L activation achieved a balanced accuracy of 73.5% and an F1-score of 0.751 (Supplementary Table 1), underscoring its relative superiority among the tested ROIs. Collectively, the ROC analysis indicates that fNIRS-measured prefrontal activation, particularly in the SFG, holds promising discriminative value (AUCs up to 0.79) for distinguishing CID patients from healthy controls. It is important to interpret these findings with nuance; while the AUC values fall in the ‘fair to good’ range and represent a significant step towards objective assessment, they currently serve as a proof of concept rather than a definitive diagnostic tool. Future studies that integrate these fNIRS metrics with other clinical and neurophysiological measures are needed to enhance predictive accuracy and validate their utility in larger, more diverse cohorts.

This study combines multi-channel fNIRS with the VFT to investigate the neural mechanisms underlying CID. By non-invasively and simultaneously assessing prefrontal cortex (PFC) activation patterns and FC, we found that insomnia patients exhibit not only reduced PFC activation during the VFT but also significantly weakened inter-channel functional connectivity. These dual abnormalities—diminished activation and disrupted connectivity—provide novel neurophysiological evidence to explain cognitive dysfunction in patients, offering greater systematic validity and explanatory power compared to previous studies focused solely on regional brain activation. The innovation of this study lies in both methodological and discovery dimensions: Methodologically, the multi-channel fNIRS approach enabled simultaneous observation of task-specific brain activity patterns (through standardized VFT paradigms) and functional connectivity networks. In terms of discoveries, we revealed that cognitive impairments in insomnia patients may originate from dual impairments in nodal functionality (local activation) and network properties (global connectivity) within the PFC neural circuitry. These findings not only refine the neuropathological model of insomnia but also identify potential dual-dimensional biomarkers for future targeted neuromodulation therapies.

Our study has several limitations that should be considered when interpreting the results. Our study has several limitations that should be considered when interpreting the results. First, the spatial resolution of fNIRS is inherently limited to the cortical surface. While this technique is excellent for assessing the prefrontal cortex, it cannot capture activity in subcortical or deep-brain regions—such as the thalamus, amygdala, or hypothalamus—which are critically involved in sleep-wake regulation and the pathophysiology of insomnia. Therefore, our findings are specific to cortical contributions and do not provide a complete picture of the neural networks involved in CID. Second, although we excluded participants with clinically significant depression (HAMD-17 ≥ 17) and anxiety (HAMA ≥ 14), we cannot rule out the potential influence of residual sub-clinical mood symptoms on our findings. Furthermore, the potential effects of medications taken by some inpatients on cerebral hemodynamics and cognitive performance were not systematically controlled for.Third, our study used a cross-sectional design, which limits our ability to determine causality. Longitudinal studies are needed to explore the temporal relationship between cortical activation, functional connectivity, and cognitive impairment in chronic insomnia. Fourth, the correlational analyses between cortical activation and sleep parameters were exploratory and were not corrected for multiple comparisons; these results should therefore be interpreted as hypothesis-generating.

Fifth, our study focused on a specific cognitive task (the VFT), which may not fully capture the range of cognitive impairment in chronic insomnia. Combining multiple cognitive tasks may provide a more complete understanding of the neurophysiological mechanisms at work. Finally, future studies should aim to recruit larger cohorts to improve statistical power and validate our results.

Conclusions

This study highlights the critical role of prefrontal and temporal cortical dysregulation, along with disrupted functional network integration, in the neuropathology of CID. The observed impairments in executive function-related brain regions and weakened interregional connectivity underscore a potential neural substrate for the cognitive and emotional disturbances commonly associated with CID. The integration of fNIRS-derived activation and connectivity metrics offers a promising framework for developing objective diagnostic tools, bridging the gap between clinical symptoms and neurobiological mechanisms.Future investigations should prioritize longitudinal designs to elucidate whether these neural alterations represent trait markers of vulnerability or consequences of prolonged sleep disruption. Expanding the scope to diverse demographic populations and incorporating multimodal neuroimaging approaches (e.g., combined fMRI and fNIRS) could enhance spatial resolution and deepen mechanistic insights. Additionally, exploring the reversibility of these neural deficits through targeted interventions—such as cognitive-behavioral therapy, pharmacotherapy, or neuromodulation—may clarify causal relationships and inform personalized treatment strategies. Addressing these questions will advance our understanding of insomnia as a brain network disorder and refine its clinical management.