Correlation of prefrontal-based resting state functional connectivity with the degree of cognitive impairment in Alzheimer’s disease: a functional near-infrared spectroscopy study

Abstract

Background

This study aims to investigate the differences in neural network connectivity within the prefrontal cortex (PFC) among elderly individuals with normal cognition (NC), mild cognitive impairment (MCI), and Alzheimer’s disease (AD) using functional near-infrared spectroscopy (fNIRS).

Results

Significant differences in functional connectivity (FC) strength were observed between the NC, MCI, and AD groups in several Brodmann areas (BA) pairs, including BA46.L-BA45.R and BA9.L-BA1.L. The most pronounced FC strength differences between the NC and MCI groups occurred at the 2nd -minute mark in BA45.R, while differences between the MCI and AD groups peaked at the 5th-minute mark in BA1.L. Additionally, the NC and MCI groups displayed FC strength differences during the first 2 minutes and first 3 minutes, again with BA45.R being central. FC strength between BA46.L-BA45.R was negatively correlated with Neuropsychiatric Inventory and Clinical Dementia Rating scores.

Conclusions

FC strength in the left dorsolateral PFC, where BA46.L and BA9.L are located, emerged as a key region for cortical dysfunction in cognitive impairment. Moreover, there were differences in FC across levels of cognitive impairment, and significant correlations between differences in FC strength in BA brain regions and cognitive level.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12993-025-00316-8.

Background

According to the World Health Organization, more than 55 million people worldwide currently live with dementia, with 10 million new cases each year. Alzheimer’s disease (AD) accounts for about 60% of these cases, with the prevalence expected to triple by 2050 [1]. As the most common neurodegenerative disease, AD poses significant challenges, particularly in terms of early diagnosis and intervention, which are crucial for slowing cognitive decline and improving patients’ quality of life. Mild cognitive impairment (MCI) is often recognized as a transitional stage between normal aging and dementia, marked by noticeable cognitive decline that still allows for a relatively high level of independence in daily activities [2, 3, 3]. The annual conversion rate of MCI to AD is estimated at 10%, with over 80% of MCI eventually progressing to AD [4, 5]. However, the subtle clinical manifestations of MCI, combined with the limitations of current diagnostic tools, often result in missed opportunities for early intervention. Therefore, improving the early diagnosis of MCI and effectively distinguishing it from AD are crucial steps in reducing the progression and incidence of AD.

Functional near-infrared spectroscopy (fNIRS) is a non-invasive assay that uses near-infrared light to penetrate the head tissue, where it is absorbed by hemoglobin in the cerebral cortex. This allows for real-time detection of changes in cortical hemodynamics, providing insights into brain activity [6, 7]. fNIRS has been widely applied in research on mood disorders, schizophrenia, and dementia [8–11]. Functional connectivity (FC) refers to the dynamic synchronization of neural activity between different regions of the brain [12–14]. Resting-state fNIRS specifically measures the correlation of neural signals across different regions during low-frequency fluctuations, making it a valuable tool for identifying brain function alterations associated with cognitive impairments [15–17]. Studies have shown that in resting-state conditions, individuals with MCI and AD exhibit altered connectivity and fluctuations in cerebral oxygenation [18–20]. Previous studies have shown that patients with MCI exhibit reduced frontotemporal and parietal cortex activation and reduced parietal low-frequency oscillations [21–23]. Further findings indicate that amnestic MCI have less activation in the frontotemporal or parietal cortex, along with altered lateral prefrontal responses [24, 25]. These observations suggest that the prefrontal cortex (PFC) is a critical region affected by AD-related impairment.

Resting-state FC (RSFC) analyses of fNIRS measures have demonstrated the potential for objectively and validly distinguishing differences in brain function between MCI and AD. This approach offers a rapid screening method for detecting MCI in clinical settings. However, A previous study reported that statistical analysis could not satisfactorily detect MCI [26]. This suggests that the differences in RSFC between MCI and AD are not yet fully established, and that whole-process RSFC analyses for screening MCI and detecting AD remain insufficient and controversial. Additionally, prior studies have often identified broader regions of potential brain injury without focusing on specific Brodmann areas (BAs). This lack of precision has hindered efforts to link specific areas of brain injury to cognitive functions. In contrast, a purely channel-based analysis, which relies on scalp coordinates and lacks direct biological annotation, is less informative than the BA framework. The BA framework anchors findings in a well-established neuroanatomical system, facilitating interpretation within the extensive human brain mapping literature and enabling direct comparisons across studies using different neuroimaging modalities. Therefore, by exploring more specific differences in RSFC between specific BA regions across cognitive levels, we can enhance the accuracy of identifying the brain regions associated with cognitive impairment. This could also help clarity the relationship between these brain regions and cognitive dysfunction.

For some patients with cognitive dysfunction, completing a full 5-minute resting-state assessment is challenging. Therefore, analyzing a single time window is more feasible in clinical practice. Dynamic RSFC analyses using different time windows, such as 20–60 second (S), provide researchers with tools to gain deeper insights into the relationship between time-varying macroscopic neural activity patterns and key aspects of cognition and behavior [27, 28]. To address the limitations of whole-process RSFC analyses and avoid false-negative results observed in previous studies, conducting more detailed RSFC analyses across different time periods is essential. This research aims to investigate how RSFC within specific BAs changes across cognitive levels during resting state and to analyze the relationship between prefrontal RSFC and cognitive function. Based on established evidence of prefrontal network disruption in neurodegeneration, the primary hypotheses are as follows: (1) There are significant differences in RSFC strength from normal cognition (NC) to MCI to AD; (2) Specific patterns of altered connectivity between distinct regions of interest (ROIs) are associated with domain-specific cognitive deficits. Additionally, this study explores the dynamic characteristics of prefrontal RSFC across different time periods, quantified using a 60S time window, under varying cognitive levels.

Methods

Participants

A total of 328 participants were finally included in the trial, including 78 in the NC group, 121 in the MCI group and 129 in the AD group. All participants underwent a thorough medical history, a neuropsychological battery, structural magnetic resonance imaging (sMRI) and routine blood tests. AD was diagnosed in accordance with the diagnostic criteria for probable AD dementia updated by the National Institute on Aging-Alzheimer’s Association (NIA/AA) in 2011 [29], and diagnosis for MCI was also based on NIA/ AA diagnostic guidelines [3]. Cognitive status was evaluated by Chinese version of Mini-Mental State Examination (cMMSE), Alzheimer’s Disease Assessment Scale-Cognitive Subscale (ADAS-Cog), Clock Drawing Test (CDT). Neuropsychiatric symptoms were assessed by Neuropsychiatric Inventory (NPI). Daily function was evaluated by Instrumental Activities of Daily Living scale (IADL). And the severity of cognitive impairment was rated by Clinical Dementia Rating scale (CDR, 0 = normal, 0.5 = questionable, 1 = mild, 2 = moderate, 3 = severe). The study strictly adhered to the Declaration of Helsinki and was approved by the Ethics Committee of The First Affiliated Hospital of Chongqing Medical University (approval number 20212901). All participants were thoroughly informed about the study procedures and written consent was obtained prior to participation.

BA Brodmann area, FC Functional connectivity, MMSE Mini-mental State Examination, ADAS-Cog Alzheimer’s Disease Assessment Scale-Cognition subscale, NPI Neuropsychiatric Inventory, CDR_SB Clinical Dementia Rating scale_Sum of Boxes score, CDR Clinical Dementia Rating scale.

Inclusion criteria

Participants with MCI: (1) objective memory problems confirmed by external sources; (2) cMMSE scores ranging from 17 to 27 for illiterate individuals, 20–27 for those with primary education, and 22–27 for those with lower secondary education; (3) CDR score of 0.5; and (4) no diagnosis of dementia.

Participants with AD: (1) presence of definite cognitive impairment meeting the criteria for dementia; (2) cMMSE adjusted for education: scores of ≤ 17 for illiterate individuals, ≤ 20 for those with secondary education, and ≤ 24 for those with university education; (3) impairment in two or more cognitive domains; (4) progressive decline in memory and other cognitive functions; (5) no impairment of consciousness.

Participants with NC: (1) no self-reported memory loss or cognitive problems; (2) normal neuropsychological performance.

Exclusion criteria

(1) An identified history of stroke; (2) history of traumatic brain injury; (3) other systemic illnesses causing cognitive impairment, such as thyroid dysfunction, syphilis, or AIDS.; (4) significant psychiatric disorders; and (5) inability to co-operate with the study procedures.

Resting state fNIRS acquisition and preprocessing

In this study, fNIRS acquisition and preprocessing were conducted in accordance with our previous study using a 48-channel system (NirScan, Dangyang Huichuang Medical Equipment Co. Ltd, China) [8]. Briefly, the 5-minute resting state recording took place in a sound-attenuated room. During this period, participants were seated comfortably and instructed to remain still with their eyes closed. 15 sources and 16 detectors were mounted on elastic caps to ensure good contact with the participant’s head. The emitters and detectors were separated by 30 mm to detect changes in oxy-hemoglobin (HbO) and deoxy-hemoglobin (HbR) using two wavelengths (730 nm and 850 nm), respectively. The channels and the corresponding brain regions are consistent with the rules of a previous study (Supplementary Tables 1 and Supplementary Fig. 1) [8]. The channel-to-ROI correspondence was cross-checked against a probabilistic atlas. For preprocessing, we used the NirSpark software package v1.7.3 (Danyang Huichuang Medical Equipment Co. Ltd, China) to analyze NIRS data. Data were preprocessed via the following steps. Motion artifacts were corrected by a moving SD and a cubic spline interpolation method. All differential path-length factors (DPF) were set to 6.0. The fNIRS signals were acquired at a sampling rate of 11 Hz. Each 60-second dynamic functional connectivity window contained 660 data points per channel. According to previous study [30], a bandpass filter with cut-off frequencies of 0.01–0.20 Hz was used to minimize noise, global trends, and biological signals (e.g., respiration and cardiac activity). The modified Beer-Lambert law was applied to convert optical densities into changes in HbO and HbR concentrations. We used HbO as our primary indicator in the following analysis because the HbO signal generally has a better signal-to-noise ratio than HbR [31].

All participants exhibited adequate signal quality after preprocessing (motion correction and filtering), as confirmed by the presence of cardiac rhythm and typical hemoglobin fluctuations in their time series. Channels with a median intensity below 20 arbitrary units were excluded, following established guidelines for channel rejection [32]. Using this criterion, an average of 2–3 channels (out of 48) per participant were identified as low-quality and removed from analysis. These were typically peripheral channels affected by hair interference or poor fit. Importantly, we required that each participant retained at least 90% of channels with good signal quality.

Brain functional connectivity

The RSFC of the different measurement channels between the groups was compared by using the false discovery rate (FDR). A 48 × 48 correlation matrix was generated for each participant. These correlation coefficients were subsequently converted to Z scores by applying Fischer’s R-Z transformation to improve normality. Differences in channel and FC were explored by analyzing data from different time windows, setting the windows at 60 s intervals or cumulatively for the first minute, the first two minutes, the first three minutes, the first four minutes, and the entire five-minute period.

Based on the correspondence between the various channels and BAs, the 48 channels were divided into 19 BA regions, which were used as the regions of interest (ROI) for this study. The specific correspondence is provided in Supplementary Tables 1 and Supplementary Fig. 1. In our analysis, each FC comparison (either at the channel-pair level or ROI-pair level) was included in the multiple comparison correction. For the channel-based analysis, this amounted to 1,128 unique channel-pair comparisons (upper triangle of the 48 × 48 matrix), and for the ROI-based analysis we had 171 unique ROI-to-ROI connections (19 ROIs) tested.

The rationale for selecting the Brodmann atlas over other available parcellation schemes was threefold. First, the Brodmann map, grounded in cytoarchitectonics, provides a historically and theoretically rich framework that is explicitly linked to cognitive functions [8, 32, 33]. This allows our findings to be directly contextualized within a vast body of literature on the neural correlates of cognitive impairment. Second, compared to broader anatomical labels (e.g., “PFC” or “frontal pole”), the BA scheme enables a more precise investigation of FC between specific subregions. This enhanced specificity is critical for achieving our aim of delineating fine-grained alterations in neural circuits associated with varying levels of cognitive performance. Finally, given that our fNIRS array was specifically designed to cover the prefrontal cortex, the Brodmann atlas is optimally suited for parcellating this limited field of view into functionally meaningful subunits. It is important to acknowledge that the BA map is not without limitations. There is known inter-individual variability in cytoarchitecture, and functional boundaries observed through neuroimaging may not perfectly coincide with the histological borders defined by Brodmann [34].

Statistical analysis

All statistical analyses were performed by IBM SPSS ver. 26.0 (IBM Corp., Armonk, USA). The NirSpark software package v1.7.3, GraphPad Prism 9, and Photoshop software were used to generate figures and graphs. For all comparisons involving multiple tests, we applied the Benjamini–Hochberg FDR correction with a q-value set at 0.05. In total, 1,128 comparisons for channel-based analysis and 171 comparisons for ROI-based analysis were corrected. Variables conforming to normal distributions were expressed as mean ± standard deviation (SD). Variables not fitting the normal distribution were analyzed using Fischer’s r-to-z transformation. One-way ANOVA was used to compare ROI-based RSFC among NC, MCI, and AD groups, and partial eta-squared(η²) was computed as an effect size indicator. Pairwise differences were further assessed with Cohen’s d effect size calculation. A two-tailed p < 0.05 was considered statistically significant. Pearson correlations were computed between neuropsychological outcomes and RSFC between ROIs, with effect sizes explicitly stated.

Results

Demographic and clinical characteristics

The results showed that no significant differences were observed in gender and age among the three groups, but there were significant differences in education level (Table 1). The lower the education level, the more severe the cognitive dysfunction, suggesting that low literacy is a risk factor for cognitive dysfunction. Neuropsychological assessments showed significant differences in MMSE, ADAS-Cog, CDT, GDS, NPI and CDR scores between the three groups (Table 1). These cognitive assessment scores showed a trend of progressive deterioration from NC to MCI to AD, suggesting that participants experienced significant declines in overall cognitive functioning, memory, learning, and executive functioning, while accompanying psycho-behavioral symptoms also became progressively more pronounced.

Table 1.

Characteristics of each group

Variables	NC (n = 78)	MCI (n = 121)	AD (n = 129)	Z or χ2	P value
Age, year	74.68 (9.26)	75.88 (7.91)	75.30 (6.70)	0.328	0.721
Female, n & %	39 (56.52%)	67 (57.26%)	77 (57.03%)	0.771	0.680
Education, year	10.95 (3.73)	10.61 (3.16)	8.71 (3.91)	5.832	0.004**
MMSE	28.73 (1.45)	23.81 (2.67)	17.71 (4.93)	160.435	<0.001***
ADAS-Cog	5.05 (3.23)	10.79 (4.11)	25.83 (9.52)	169.071	<0.001***
CDT	13.08 (2.06)	11.41 (2.68)	7.54 (3.26)	62.084	<0.001***
GDS	2.42 (2.18)	4.90 (2.62)	5.55 (4.29)	11.582	<0.001***
NPI	0.28 (1.33)	3.48 (4.58)	8.86 (7.83)	40.493	<0.001***
CDR
Global Score	0.00 (0.00–0.00)	0.50 (0.50–0.50)	0.50 (0.50-1.00)	119.763	<0.001***
SB	0.25 (0.00-0.50)	1.50 (1.00-2.50)	4.50 (3.50–5.63)	214.095	<0.001***

Hemodynamic changes in the PFC

There was a difference in the real-time HbO changes in the different channels of the three groups (Fig. 1A). The HbO changes in the channels gradually weakened from the NC and MCI groups to the AD group. There were also differences in HbO between the different regions of the PFC in the three groups (Fig. 1B). The change and standard deviation of the mean HbO concentration in the NC, MCI, and AD groups were respectively − 0.003817 ± 0.005711, − 0.001998 ± 0.004758, − 0.001155 ± 0.003250. Compared to the NC group, the AD group maintained a higher level of mean HbO concentration. In addition, there were no differences in mean HbO concentrations between the NC and MCI groups, as well as between the MCI and AD groups (Fig. 1C).

Fig. 1.

Differences in mean HbO concentration and channel change curves for the three groups. A Curves of HbO and HbR under each channel in each group. B 3D results of HbO oximetry curves under each channel in each group. C Statistics on the results of HbO oximetry curves under each channel in each group. NC normal cognition; MCI mild cognitive impairment; AD Alzheimer’s disease; HbO oxy-hemoglobin; HbR deoxy-hemoglobin

Channel-based functional connectivity among groups

The mean RSFC strength was sequentially lower in the NC, MCI, and AD groups, with mean and standard deviation of 0.35809 ± 0.10895 in the NC group, 0.35397 ± 0.11097 in the MCI group, 0.33919 ± 0.1099 in the AD (Fig. 2A). At the same time, compared with the NC and MCI groups, the difference in RSFC strength between different BAs in the AD group was also the smallest, and the RSFC strength of the BA in the AD group was significantly weakened (Fig. 2B).

Fig. 2.

The FC situation in three groups in intra-group. A Maps of average FC matrices for all groups. B The BA differences between the three groups in 5 minutes. NC normal cognition, MCI mild cognitive impairment, AD Alzheimer’s disease, FC Functional connectivity, BA Brodmann area

BA-based functional connectivity

All 48 channels were divided into different BAs to further validate the connectivity characteristics among BAs. Key lesion areas for MCI and AD were extracted by ANOVA test and FDR correction. There were 2 connections with significant between-group differences. Within 5 min, there were differences in a total of 4 ROIs, including BA46.L-BA45.R and BA9.L-BA1.L. The FC strength of BA9.L-BA1.L weakened in the NC group, MCI group, and AD group. The FC strength of BA46.L-BA45.R was the highest in the NC group (Fig. 3A and Supplementary Table 2). And FC kept changing between the three groups in different time windows (Fig. 3B-F and Supplementary Table 3). Furthermore, the correlation analysis revealed that the FC strength of BA46.L-BA45.R was negatively correlated with NPI-Delusion, CDR-Direction and CDR-Social (Table 2).

Fig. 3.

The differences in BA across time windows in the three groups. A Three groups of BA connections and statistics in the 0–300 S window. B Three groups of BA connections and statistics in the 0–60 S window. C Three groups of BA connections and statistics in the 60–120 S window. D Three groups of BA connections and statistics in the 120–180 S window. E Three groups of BA connections and statistics in the 180–240 S window. F Three groups of BA connections and statistics in the 240–300 S window. NC normal cognition, MCI mild cognitive impairment, AD Alzheimer’s disease, FC Functional connectivity, BA Brodmann area

Table 2.

Correlation between different BA FC strength and neuropsychological assessment scores

Sub-region interaction	BA46.L-BA45.R			BA9.L-BA1.L
	r	p	q	r	p	q
MMSE	0.166	0.009	0.099	0.177	0.005	0.055
ADAS-Cog	− 0.158	0.013	0.072	− 0.156	0.014	0.051
NPI	− 0.101	0.113	0.113	− 0.127	0.046	0.101
NPI-Delusion	− 0.156	0.033	0.045*	− 0.027	0.715	0.715
NPI-Anxiety	− 0.182	0.014	0.051	− 0.089	0.234	0.322
CDR_SB	− 0.140	0.029	0.053	− 0.119	0.064	0.117
CDR-Memory	− 0.144	0.032	0.050	− 0.169	0.012	0.066
CDR− Direction	− 0.140	0.039	0.048*	− 0.115	0.091	0.143
CDR-Solving	− 0.124	0.067	0.074	− 0.155	0.022	0.061
CDR-Social	− 0.156	0.022	0.048*	− 0.060	0.382	0.467
CDR-Housework	− 0.158	0.019	0.052	− 0.030	0.656	0.722

Outcomes of channel-based functional connectivity in 60s-time window

Compared to the other two groups, the NC-MCI groups had the highest number of differential channels and BAs between them within 2nd minute, whereas the channels and BAs between NC-AD and MCI-AD were more evenly distributed on the left and right sides, and the number of discrepant channels was decreasing (Fig. 4A, and Supplementary Fig. 2A). To further confirm the location of the most sensitive cortex, we summarized the channel locations, and the highly interconnected region of the channel between NC and MCI was BA45.R, which was mainly concentrated in the right PFC (Fig. 3C). In contrast, compared with the other two groups, the number of differential channels between MCI and AD was the highest, with the highly interconnected region being BA1.L in the left PFC and temporal lobe within 5th minute (p< 0.05) (Fig. 4B and Supplementary Fig. 2B), while the number of differential channels between the NC-MCI group and MCI-AD was lower (Fig. 3F). The FC situation between the ROIs in each of the three groups (NC, MCI, and AD) at different time windows is also different (Supplementary Tables 3 and Supplementary Fig. 4–7). It should be noted that these represent static connectivity within a 5-minute period, and the variability between moments has not been quantified, which may be addressed in future research.

Fig. 4.

Intergroup analysis comparison of FC strength differences between different BAs at 2nd-minute and 5th-minute intervals. A The differences of BAs FC strength between the three groups over 60–120 s. B The differences of BAs FC strength between the three groups over 240–300 s. NC normal cognition, MCI mild cognitive impairment, AD Alzheimer’s disease, FC Functional connectivity, BA Brodmann area

Outcomes of channel-based functional connectivity in different cumulative time windows

Within the first 2 and 3 minutes, 47 and 50 channel FCs were significantly different between groups more evenly distributed between the right and left hemispheres of the prefrontal and temporal lobes, respectively (Fig. 5A, Supplementary Fig. 3A). Interestingly, the NC-MCI groups had the highest number of differential channels between them compared to the other two groups. And in further confirming the location of the most sensitive cortex, it was found that the channels between NC and MCI were all mainly concentrated in the right prefrontal lobe at the two different times mentioned above, and the highly interconnected region were BA45.R (Fig. 5B and Supplementary Fig. 3B).

Fig. 5.

Intergroup analysis comparing differences in FC strength between different BAs in the first 2 minutes and the first 3 minutes. A Differences in BA within the first 2 minutes between the three groups. B Differences in BA within the first 3 minutes between the three groups. NC normal cognition, MCI mild cognitive impairment, AD Alzheimer’s disease, BA Brodmann area, FC Functional connectivity

Discussion

This study focuses on extracting sensitive connections at both the channel and ROI levels to explore prefrontal hemodynamic and FC variability in NC, patients with MCI and with AD using resting-state fNIRS data. Notably, significant differences in FC strength were observed between the NC, MCI, and AD groups in several BA pairs, including BA46.L-BA45.R and BA9.L-BA1.L. The most pronounced FC strength differences between the NC and MCI groups occurred at the 2nd -minute mark, with BA45.R identified as a highly interconnected region, while differences between the MCI and AD groups peaked at the 5th -minute mark, with BA1.L as the highly interconnected region. Additionally, the NC and MCI groups displayed significant FC strength differences in the prefrontal during the first 2 minutes and first 3 minutes, again with BA45.R being central. These findings enhance our understanding of the prefrontal network’s involvement in varying degrees of cognitive impairment.

Elevated HbO in AD echoes neurovascular compensation for synaptic loss. The neural inefficiency model points to a decrease in processing efficiency in the PFC with age, possibly due to underlying pathological changes such as reduced PFC gray matter volume, altered white matter connectivity, and neurochemical alterations [35]. Whereas there is a paradoxical relationship between the structure and function of the PFC, in that the PFC is the region with the greatest volume reduction [36], it is also the most common site of overactivation in response to inefficiencies in the PFC region, which may or may not be compensatory [37]. The potential of fNIRS as a sensitive tool for detecting early functional changes requires validation against gold-standard biomarkers in future studies. Furthermore, it remains unclear whether these changes reflect compensatory neural activity in individuals with impaired cognitive function [36, 37] or reduced synaptic efficiency, which demands higher metabolic costs for equivalent tasks, leading to neural inefficiency. This issue warrants further investigation. Additionally, no differences were observed between the NC and MCI groups, and the reasons for this finding require further exploration.

In patients with MCI and AD, reduced communication within PFC has been associated with clinical symptoms such as executive function, attention, visuospatial function and memory [38–41]. Previous studies have also shown that FC strength is significantly reduced in patients with MCI [42, 43]. Significant variations in FC strength are observed across distinct BAs of the bilateral PFC among NC, MCI, and AD patients. Specifically, FC strength of BA46.L-BA45.R and BA9.L-BA1.L is positively correlated with the degree of cognitive impairment. These BAs involve the left dorsolateral PFC and Broca’s area, which are closely related to language function and memory function. Reduced communication between these brain regions leads to decreased language function and memory function, which is consistent with the clinical symptoms of patients with cognitive impairment. In addition, BA46.L-BA45.R is negatively correlated with NPI-Delusion scores, and these BA regions involved the left dorsolateral PFC. Among cognitive disorders, delusion is often one of the major NPI clinical symptoms, causing serious burden and distress to patients and families. However, psychotropic drugs have adverse effects on cognitive function, and clinical caution is exercised regarding the use of psychotropic drugs. In practice, the decision of whether to use them is more often based on the patients’ symptomatic manifestations and scales, while the present study confirmed the correlation between BA46.L-BA45.R and NPI-Delusion, which suggests that it may be important to assist in the decision of whether or not to use psychotropic medications by detecting the FC strength of BA46.L-BA45.R. Above all, the FC patterns in these regions show potential as a complementary tool to support clinical diagnosis, pending validation in independent cohorts. Future studies should integrate fNIRS-RSFC with neuroimaging, fluid biomarkers, and digital phenotyping to develop clinically actionable diagnostic algorithms and definitive diagnostic markers for fNIRS.

The process of FC change in AD patients has been less observed in the published studies on RSFC, while our study finds that the FC network is in a constantly changing throughout the resting state. This study indicates that the right PFC of the brain in the MCI group differed significantly from the NC group in the early stages of FC and less from AD, indicating that a certain site and degree of cognitive impairment had already occurred in the MCI stage. However, with the increase of time, the difference of the corresponding FC of the MCI group and the NC group gradually decreased, and the left PFC appeared significantly different from the AD group, indicating that compensatory activation could occur in MCI and better cognitive function was maintained compared with AD. Moreover, our findings demonstrated significant temporal variability in FC across cognitive impairment stages. This suggests the existence of different resting-state network patterns in the prefrontal lobe of the brain during the progression of cognitive function from NC to MCI to AD. Furthermore, time-resolved clustering of fNIRS-based dynamic RSFC can effectively identify and extract dominant functional networks. Research has found that the frontal-parietal-temporal network, occipital network, and sensorimotor network are highly consistent across different sliding time window lengths of 20, 30, and 60 s [27]. The accuracy of the fNIRS assay could be improved by increasing the RSFC of the 1-minute time window analyzed. At the same time, Previous study have found that 1 minute of fNIRS scan data was sufficient to generate stable and highly similar FC patterns compared to the full 10 minutes, which supports the fact that reliable connectivity estimates can be obtained in a 60-second window [44]. Not only that, for some patients who are unable to cooperate in completing the entire 5-minute resting state procedure, a 1-minute fNIRS data acquisition can increase the success of the test. Therefore, examining RSFC differences in the 1-minute time window across cognitive levels is important to increase the accuracy of the 5-minute fNIRS and the clinical feasibility of the fNIRS. Although a 60-second window can capture dynamic interactions, FC estimates derived from shorter time windows may exhibit increased sensitivity to noise and reduced stability compared to longer periods, particularly at lower frequencies of the resting-state fluctuation spectrum. Moreover, these time-specific results should be confirmed in an independent dataset to fully avoid any potential circular inference. Finally, the analysis in this study did not characterize rapidly changing connectivity, and therefore may have missed subtle dynamic patterns that overlapping window or dynamic graph analysis might have revealed. Apply sliding window correlation or time-varying network metrics in subsequent studies.

Several limitations should be noted. First, although we excluded participants with major vascular comorbidities, we did not fully account for other potential confounders, such as depressive symptoms, specific medication use (e.g., antidepressants, benzodiazepines), or subclinical vascular risk factors (e.g., hypertension). Secondly, while our group matching and primary adjustments for age and sex align with standard practices in fNIRS research [45], we cannot entirely rule out their potential impact on prefrontal hemodynamics. Future studies with larger sample sizes should systematically control for these factors to confirm the specificity of our findings to AD pathology. Third, longitudinal studies with follow-up confirmation of control group stability are essential to validate and potentially strengthen the effects reported here. This is particularly critical, as recent research suggests that cohorts of cognitively normal individuals may include those with preclinical AD pathology who may subsequently progress to MCI [46]. Such designs would be crucial for determining the predictive value of fNIRS for identifying individuals at the preclinical stage. The lack of comparison with other dementias is a key limitation that precludes claims of specificity. Future studies directly comparing AD with other neurodegenerative conditions are the essential next step.

Conclusion

In summary, the present study demonstrates that different cognitive levels have different RSFC networks, and that the left dorsolateral PFC may be a key region for cortical impairment in patients with cognitive impairment. There was significant correlation between FC strength difference in BA brain region and cognitive impairment. These findings suggest that FC metrics show promise for detecting early cognitive impairment, but these findings remain preliminary. Replication in larger, independent cohorts with longitudinal designs is essential to establish clinical utility.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AD

Alzheimer’s disease

ADAS-Cog

Alzheimer’s Disease Assessment Scale-Cognitive Subscale

BA

Brodmann area

CDR

Clinical Dementia Rating scale

CDT

Clock Drawing Test

FC

Functional connectivity

fNIRS

Functional near-infrared spectroscopy

HbO

Oxy-hemoglobin

HbR

Deoxy-hemoglobin

IADL

Instrumental Activities of Daily Living scale

MCI

Mild cognitive impairment

MMSE

Mini-Mental State Examination

NC

Normal cognition

NPI

Neuropsychiatric Inventory

PFC

Prefrontal cortex

ROI

Regions of interest

Author contributions

YL and WHY designed the study. MC and WBZ analyzed data and wrote the paper. YL and WHY revised the manuscript. MC, WBZ, MYY, FXZ, JQS, YMG, YXC and QT collected the data and assisted with writing the article.

Funding

This work was supported by grants from Chongqing Talent Plan (grant number cstc2022ycjh-bgzxm0184); Science Innovation Programs Led by the Academicians in Chongqing (grant number 2022YSZX-JSX0002CSTB); Key Project of Technological Innovation and Application Development of Chongqing Science & Technology Bureau (grant number CSTC2021jscx-gksb-N0020); Program for Youth Innovation in Future Medicine, Chongqing Medical University (grant number W0166); Chongqing Medical Key Discipline and Regional Medical Key Discipline Development Project (grant number 0201[2023] No.160 202412); The First Affiliated Hospital of Chongqing Medical University 2022 Chongqing Research Innovation Program for Graduate Student (CYB22199) (grant number CYYY-BSYJSCXXM-202334) and Chongqing Medical Key Discipline and Regional Medical Key Discipline DevelopmentProject 0201[2022]No. 144 202325zdxk202105.

Data availability

The datasets analysed during the current study are not publicly available due data confidentiality obligations but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of The First Affiliated Hospital of Chongqing Medical University (approval number 20212901) and has been performed in accordance with the ethical standards laid down in the Declaration of Helsinki and its later amendments. This study did not involve any intervention with the participants. It mainly analyzed the changes in HbO and HbR in the participants’ brains and used cognitive scales to assess the participants’ cognitive function.

Consent for publication

All participants were informed in detail about the study procedures and obtained written consent prior to their participation in the study, and consented to publication.

Competing interests

The authors declare no competing interests.