fNIRS as a biomarker for individuals with subjective memory complaints and MCI

Abstract

INTRODUCTION

Identifying individuals at risk of developing dementia is crucial for early intervention. Mild cognitive impairment (MCI) and subjective memory complaints (SMCs) are considered its preceding stages. This study aimed to assess the utility of functional near‐infrared spectroscopy (fNIRS) in identifying individuals with MCI and SMC.

METHODS

One hundred fifty‐one participants were categorized into normal cognition (NC); amnestic MCI (aMCI); non‐amnestic MCI (naMCI); and mild, moderate, and severe SMC groups. Task‐related prefrontal hemodynamics were measured using fNIRS during a visual memory span task.

RESULTS

Results showed significantly lower oxyhemoglobin (HbO) levels in aMCI, but not in naMCI, compared to the NC. In addition, severe SMC had lower HbO levels than the NC, mild, and moderate SMC. Receiver operating characteristic analysis demonstrated 69.23% and 69.70% accuracy in differentiating aMCI and severe SMC from NC, respectively.

DISCUSSION

FNIRS may serve as a potential non‐invasive biomarker for early detection of dementia.

Highlights

• Only amnestic mild cognitive impairment (aMCI), but not non‐amnestic MCI, showed lower oxyhemoglobin (HbO) than normal individuals.

• Reduced HbO was observed in those with severe subjective memory complaints (SMCs) compared to normal cognition (NC), mild, and moderate SMCs.

• Functional near‐infrared spectroscopy measures were associated with performance in memory assessments.

• Prefrontal hemodynamics could distinguish aMCI and severe SMC from NC.

1. BACKGROUND

Dementia affects > 55 million people globally ¹ and poses significant burdens for individuals, caregivers, and society. ² Recent advancements in pharmacological interventions for early‐stage dementia ³ , ⁴ have highlighted the importance of identifying early signs of dementia. Mild cognitive impairment (MCI) is a transitional stage between normal aging and dementia. ⁵ Individuals with MCI exhibit lower performance on neuropsychological assessments while maintaining independent living abilities. ⁶ The global prevalence of MCI among individuals aged ≥ 50 years is ≈ 15.56%, ⁷ with a higher conversion rate to dementia (9.9% after 2 years) compared to normal cognition (NC; 1.0%). ⁸ MCI can be categorized into amnestic MCI (aMCI), characterized by memory impairment, and non‐amnestic MCI (naMCI), characterized by impairments in cognitive domains other than memory. ⁹ , ¹⁰ , ¹¹ aMCI is more predictive of Alzheimer's disease (AD), while naMCI indicates other dementia subtypes. ¹²

Subjective memory complaint (SMC) has emerged as a potentially preclinical stage of dementia, preceding MCI. ¹³ , ¹⁴ , ¹⁵ It refers to an individual's self‐perceived cognitive decline compared to their previous functioning. ¹⁶ , ¹⁷ SMC affects ≈ 50% of individuals aged 50 to 59, and 63% of those aged 80 to 100. ¹⁸ Intriguingly, individuals with SMC, despite not meeting the MCI criteria, have a 2.17 times higher risk of developing dementia compared to individuals without SMC. ¹⁷ They also exhibit an increased risk of abnormalities in dementia‐related biomarkers, regional brain hypometabolism, and atrophy in the medial temporal lobe, compared to individuals with NC. ¹⁷ , ¹⁹ , ²⁰ , ²¹

While the diagnosis of MCI and dementia is still based heavily upon clinical diagnosis using standardized neuropsychological tests and clinical interviews, much research has focused on identifying biomarkers for detecting the preclinical stage of dementia. The National Institute on Aging and the Alzheimer's Association recently updated their diagnostic guidelines in 2023. ²² They proposed that AD should be defined biologically rather than based on clinical syndromes. They suggested diagnosing the disease in vivo using abnormalities in core biomarkers such as positron emission tomography (PET) and fluid from cerebrospinal fluid (CSF) and plasma. These biomarkers include A (amyloid beta), and T (tau) changes that PET can detect, which are crucial for staging the disease. ²³ Despite concerns raised by this approach, ²⁴ , ²⁵ the need for biomarker development is warranted. However, the invasive nature of PET poses challenges, leading to low acceptance among older adults in the preclinical stage. ²⁶ The present study aims to investigate the potential of using functional near‐infrared spectroscopy (fNIRS) as a non‐invasive and cost‐effective method to identify individuals with MCI and SMC.

fNIRS is a non‐invasive and portable method for measuring oxyhemoglobin (HbO) and deoxyhemoglobin (HbR) levels in cerebral blood flow ²⁷ and has shown promise in investigating cognitive function in individuals with NC, MCI, and dementia. ²⁸ Studies using fNIRS have revealed decreased cognitive task‐related HbO changes in the prefrontal brain regions of individuals with MCI and dementia compared to those with NC. ²⁹ Lower HbO during demanding cognitive tasks in individuals with SMC has been associated with lower memory test performance, ²⁸ suggesting fNIRS holds the potential for differentiating memory abilities in SMC individuals. The reduced frontal activation is more severe in dementia than in MCI, ²⁸ and more pronounced in MCI compared to SMC, ³⁰ indicating that decreased frontal HbO changes may serve as a biomarker for dementia progression.

The present study aimed to examine the sensitivity of fNIRS in differentiating NC and aMCI, as well as NC and naMCI. Furthermore, this study compared task‐related activation in normal older adults with varying degrees of SMC. This evaluation of fNIRS provided insights into its potential to differentiate individuals in the early stages of dementia, including the preceding stage of MCI. It was hypothesized that the aMCI group would show lower levels of HbO than the NC group during the computerized memory task. Similarly, individuals with more severe SMC are expected to exhibit lower HbO levels than the NC group.

RESEARCH IN CONTEXT

1. Systematic review: A literature search in databases (e.g., PubMed, Google Scholar) revealed a gap in research on using functional near‐infrared spectroscopy (fNIRS) to differentiate individuals with amnestic mild cognitive impairment (aMCI) and non‐amnestic MCI from normal individuals. Additionally, there is limited understanding of prefrontal hemodynamics in normal individuals with varying degrees of subjective memory complaints (SMCs).

2. Interpretation: The present study identified prefrontal hemodynamics, measured by fNIRS, as a potential biomarker for the early stages of dementia and its preclinical stage. The study also found a similar activation pattern in individuals with aMCI and those with SMC, which supports the notion that SMC can be considered a preclinical stage of dementia.

3. Future directions: The use of fNIRS as a biomarker for SMC and MCI can be evaluated longitudinally with larger populations. Additionally, future studies have the potential to combine markers of cerebral blood flow with other pathological biomarkers to enhance our understanding of the progression of cognitive decline.

2. METHODS

2.1. Participants

The participants were initially assessed to determine whether they met the criteria for MCI based on Jak/Bondi's criteria. ³¹ This assessment involved administering tests in three cognitive domains: memory, language, and attention/executive function (Table 1). Participants were classified as having MCI if they obtained impaired scores on these tests, which were specifically defined as scores falling more than one standard deviation below the age‐ and education‐corrected normative mean on at least two neuropsychological measures within the same cognitive domain, or impaired scores in all three cognitive domains. Among those classified as having MCI, further subgroup classifications were made. Specifically, participants with aMCI were identified as having impaired scores in the memory domain, while participants with naMCI were identified as having impaired scores in either the language, or attention/executive function domain, or both, but not in the memory domain.

TABLE 1.

Neuropsychological measures for assessing different cognitive domains.

Memory	Speed/executive functions	Language
HKLLT 10‐minute delayed recall	STT‐A completion time	CF (animal) total unique words in 60 seconds
HKLLT 30‐minute delayed recall	STT‐B completion time	CF (transportation) total unique words in 60 seconds
HKLLT recognition	FPT unique design	BNT spontaneous naming

Note: BNT, Boston Naming Test; CF, Category Fluency; FPT, Five‐Point Test; HKLLT, Hong Kong List Learning Test; STT, Shape Trail Test.

The remaining participants who did not meet the MCI criteria were further classified as NC participants. These participants were further divided based on their SMCs, measured by the Abbreviated Memory Inventory for Chinese (AMIC). ³² The AMIC is a five‐item self‐reported questionnaire that assesses everyday memory complaints, such as “forgetting where things are placed” or “unable to recall the names of good friends.” Participants received a score of one for each endorsed memory complaint item, resulting in an AMIC score ranging from zero to five. Participants with a score lower than two were classified as not having significant SMC (i.e., NC). ³² Participants with SMC were divided into subgroups based on the severity of their complaints: mild SMC (a score of three), moderate SMC (a score of four), and severe SMC (a score of five). As a result of this classification process, there were a total of 37 participants with NC, 24 participants with mild SMC, 33 participants with moderate SMC, 29 participants with severe SMC, 15 participants with aMCI, and 13 participants with naMCI.

It is noted that all 151 recruited participants scored 21 or above on the Hong Kong version of the Montreal Cognitive Assessment, which is above the cutoff score for identifying dementia. ³³ This indicates the absence of dementia among the participants.

2.2. Procedures

All participants underwent a standardized neuropsychological assessment. The assessment consisted of tests for assessing memory, executive function, attention, language, and global cognitive function. These tests included the Hong Kong List Learning Test (HKLLT), ³⁴ which is a widely used verbal memory test in Hong Kong assessing the memory abilities of older adults, ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ as well as the Shape Trail Test, ⁴⁰ a culturally unbiased version of the Trail‐Making Test. In addition, the Five‐Point Test, ⁴¹ the Boston Naming Test, ⁴² the Category Fluency Test, ⁴³ and the Hong Kong version of the Montreal Cognitive Assessment ³³ were administered. Subsequently, an fNIRS session followed. During the fNIRS session, participants performed a visual memory span task while their prefrontal hemodynamic activity was recorded. Participants were given instructions to remain stationary and minimize any head and body movement to prevent any unnecessary motion artifacts.

2.3. Visual memory span task

The visual memory span task used in this study was adapted from a previous fNIRS study that investigated the prefrontal hemodynamics of older adults. ⁴⁴ Each trial of the visual memory span task began with a 10‐second control task period, during which participants were instructed to focus their attention on a fixation cross displayed at the center of the computer screen. After the control task, nine blue square blocks were presented on the screen for 1 second. Subsequently, the blocks transitioned from blue to yellow in a sequential manner, changing color every second. Participants were instructed to memorize the sequence in which the blue blocks transformed into yellow. After the encoding period, a retrieval phase followed. During this phase, a “start” cue appeared in the upper right corner of the screen, accompanied by a “finish” button positioned in the lower right corner. Participants were required to reproduce the sequence by selecting the square blocks on the screen in the same order as they were originally presented. They had to click the “finish” button to complete their responses (Figure 1A). The task comprised 18 trials, with each span sequence consisting of 2 trials. The span sequences varied in length, ranging from two blocks to nine blocks (Figure 1B). Before the main task, participants underwent two practice trials to familiarize themselves with the task. Stimulus presentation was conducted using PsychoPy version 2022.2.4. ⁴⁵ The visual memory span was calculated based on the longest sequence length that participants accurately reproduced in at least any one trial out of the two trials for each span sequence. It is noted that the task continued even after participants reached their longest correct span length.

FIGURE 1.

Flow diagram showing (A) a trial with the span level 2, and (B) the flow of the visual memory span task. (C) Positions of the fNIRS optodes and measurement channels. The red and blue dots represent the source and detector probes, respectively. The yellow lines represent the 16 fNIRS measurement channels. fNIRS, functional near‐infrared spectroscopy.

The visual memory span task used in this study is based on the Corsi Block‐Tapping Test, which originated in the early 1970s. ⁴⁶ It has been a common test to assess visual spatial memory in research and clinical settings. ⁴⁷ , ⁴⁸ , ⁴⁹ Although n‐back task is a relatively more common test in neuroimaging studies (i.e., functional magnetic resonance imaging and fNIRS), we chose to use the visual memory span task for two reasons: First, in our experience with previous studies, some older adults, especially those with low levels of education, have difficulty understanding the complex requirements of n‐back tasks. The visual memory span task instruction is relatively easy for older adults to understand. Therefore, using the visual memory span task will increase the validity of the results, reducing the possibility that poor performance is due to not understanding the instruction. Second, the visual memory span task has nine levels, while the n‐back task usually only has three levels. Our previous studies showed that the gradually increasing difficulty level of the visual memory span task makes it a more sensitive measure of hemodynamic change associated with effort. ⁴⁴ , ⁵⁰ , ⁵¹

2.4. Hemodynamic response

To measure prefrontal hemodynamic activity during the visual memory span task, we used a 16‐channel OEG‐SpO2 system (Spectratech Inc.). This system uses near‐infrared light with wavelengths of 770 and 840 nm to estimate the relative concentration of HbO in the participants’ prefrontal cortex, using the modified Beer–Lambert Law. The fNIRS system consisted of six sources and six detectors arranged in a 2 ×  6 matrix configuration (Figure 1C), with a separation distance of 3 cm between the sources and detectors. The center of the bottom probe was positioned approximately on FpZ, following the international 10/20 system. The OEG‐SpO2 system had a sampling rate set at 12.21 Hz.

2.5. fNIRS data preprocessing

The fNIRS raw intensity data were processed and converted into the HomER3 data format ⁵² for subsequent preprocessing. First, the negative intensity values resulting from noisy data were corrected using the hmrR_PreprocessIntensity_Negative function, and the hmrR_PruneChannels function was used to remove noisy channels with a signal‐to‐noise ratio < 10 dB. On average, 1.93% (standard deviation [SD] = 5.76%) of the channels were removed. Next, the intensity signal was converted to optical density changes using the Intensity2OD function. Then, a low‐pass filter with a cutoff frequency of 0.1 Hz was applied to eliminate high‐frequency noise using the hmR_BandpassFilt function. The filtered optical density data were further transformed into changes in relative HbO and HbR using the modified Beer–Lambert law implemented in the hmR_OD2Conc function. A default differential pathlength factor of six was used for both wavelengths. To improve the signal quality and reduce noise, a correlation‐based signal improvement (CBSI) technique was applied using the hmrR_MotionCorrectCbsi function. The CBSI‐corrected HbO and HbR data were then baseline‐corrected using the data recorded during the 10‐second control task period before the start of the task. This baseline correction was performed using the hmrR_BlockAvg function. After the baseline correction, the data were averaged across all time points within each trial and then across the two trials for each span length. The HbO was then averaged across for analysis. Because the baseline‐corrected HbR signal mirrors the HbO signal after CBSI correction, only the HbO data were considered for further analysis.

2.6. Data analysis

Analysis of variance (ANOVA) and chi‐square tests were used to examine the between‐group differences in terms of demographic variables, neuropsychological assessment performance, and the behavioral performance of the visual memory span task. For the HbO data, repeated measures ANOVA were used to evaluate the interaction between group (NC and different MCI subgroups, or NC and different SMC subgroups) and span level (from span level 2 to level 9). To further identify specific group differences, post hoc comparisons were performed. In addition, stepwise discriminant analysis was conducted to assess the discriminatory abilities of HbO data in differentiating group memberships. Furthermore, stepwise binary logistic regression analysis was performed to determine the discriminatory abilities of HbO data in differentiating between the NC group and other subgroups such as aMCI or severe SMC. This was followed by receiver operating characteristic (ROC) analysis. Furthermore, Pearson correlation coefficients were computed to evaluate the association between the objective memory test scores and the HbO at all span levels among NC with different degrees of SMC. To identify potential patterns and associations without prematurely dismissing any potentially meaningful findings, corrections for multiple comparisons were not performed. However, an effect size was included to provide an estimation of the magnitude of between‐group and within‐group differences. For effect size estimation, Cohen d and partial eta squared (${\eta }_p^2$) were used. The statistical analyses were conducted using SPSS 28.0 (IBM Corporation). The significance level of all tests was set at 0.05 (two‐tailed).

3. RESULTS

3.1. Comparison between NC and two MCI subgroups

3.1.1. Demographic information and neuropsychological assessment performance

Table 2 presents the demographics and neuropsychological assessment performance of the NC, aMCI, and naMCI groups. The demographic variables showed no significant differences among the three groups, P = 0.11 to 0.64. In terms of neuropsychological assessment performance, the NC group demonstrated significantly better performance compared to either or both MCI subgroups in almost all neuropsychological measures, P ≤ 0.031. Specifically, the NC group exhibited higher scores than the aMCI group in the HKLLT and in the transportation condition of the Category Fluency Test, P < 0.001. Regarding the naMCI group, the NC group displayed significantly faster response time in the Shape Trail Test trial B, as well as significantly higher scores in the Five‐Point Test, both the animal and transportation conditions of the Category Fluency Test, the Boston Naming Test, and the Hong Kong version of the Montreal Cognitive Assessment, P ≤ 0.016.

TABLE 2.

Demographic information and neuropsychological assessment performance of individuals with normal cognitive function, aMCI, and naMCI.

	NC (n = 37)		aMCI (n = 15)		naMCI (n = 13)
Variables	M	SD	M	SD	M	SD	F/χ2	p
Demographics
Age (year)	65.39	6.49	67.11	4.03	66.04	8.75	0.37	0.69
Sex (F/M)	24/13		7/8		11/2		4.39	0.11
Education (year)	13.59	3.97	13.67	3.58	11.85	4.38	1.04	0.36
Hong Kong List Learning Test
Total learning	29.30	5.08	20.00	5.95	26.31	6.91	14.33	<0.001	NC***, naMCI** > aMCI
10‐minute delayed recall	11.08	2.47	4.80	2.18	10.23	3.22	32.61	<0.001	NC***, naMCI*** > aMCI
30‐minute delayed recall	11.11	2.46	4.40	1.99	9.62	3.18	37.92	<0.001	NC***, naMCI*** > aMCI
Recognition	88.34	10.74	70.42	19.97	89.42	12.60	10.22	<0.001	NC***, naMCI*** > aMCI
Category Fluency Test
Animal	19.35	4.81	17.13	2.67	13.31	3.75	10.01	<0.001	NC***, aMCI* > naMCI
Transportation	15.22	3.88	11.47	2.45	10.85	1.95	12.12	<0.001	NC > aMCI***, naMCI***
Boston Naming Test
Spontaneous naming	25.62	2.45	25.60	1.84	22.77	3.70	6.09	0.004	NC***, aMCI** > naMCI
Shape Trail Test
Trail A completion time (s)	42.75	13.00	45.80	18.15	49.98	17.79	1.11	0.34
Trail B completion time (s)	98.39	26.98	111.37	38.98	122.09	29.89	3.16	0.049	naMCI > NC*
Five‐Point Test
Total unique design	28.78	9.20	25.60	8.84	16.15	4.52	10.85	<0.001	NC***, aMCI** > naMCI
HK‐MoCA	27.51	2.34	26.40	2.56	25.38	3.04	3.66	0.031	NC > naMCI*

Abbreviations: aMCI, amnestic mild cognitive impairment; HK‐MoCA, Hong Kong version of Montreal Cognitive Assessment; naMCI, non‐amnestic mild cognitive impairment; NC, normal cognition; SD standard deviation.

^* P < 0.05, ^** P < 0.01, ^*** P < 0.001.

3.1.2. Behavioral performance of visual memory span task

The behavioral performance, which represents the longest sequence accurately reproduced by participants, was measured and compared. The NC, aMCI, and naMCI groups achieved average scores of 5.86 (SD = 1.32), 5.07 (SD = 0.59), and 5.08 (SD = 1.04), respectively (Figure 2). The ANOVA results indicated a significant difference among these three groups, F(2,62) = 3.88, P = 0.026, ${\eta }_p^2$ = 0.11. Post hoc comparisons revealed that the NC group exhibited significantly higher scores compared to both the aMCI group, P = 0.025, d = 0.69, as well as the naMCI group, P = 0.036, d = 0.63, while no significant difference was observed between the aMCI and naMCI groups, P = 0.98.

FIGURE 2.

Visual memory span score between individuals with NC, aMCI, and naMCI. The error bars represent ±  one standard error of the mean. The asterisks represent the significance levels of group differences in score. ^* P < 0.05. aMCI, amnestic mild cognitive impairment; naMCI, non‐amnestic mild cognitive impairment; NC, normal cognition.

3.1.3. Hemodynamic response during the visual memory span task

The analysis of HbO levels during the visual memory span task among the three groups revealed a significant group × span interaction, with F(14,112) = 1.91, P = 0.033, ${\eta }_p^2$ = 0.019 (Figure 3A,B), while the main effect of group was not significant, F(2,62) = 0.96, P = 0.39, ${\eta }_p^2$ = 0.030. Post hoc comparisons indicated a significant difference in HbO levels between the NC and aMCI groups at span levels 8 (P = 0.041, d = 0.62) and 9 (P = 0.033, d = 0.64). Specifically, at span level 8, the NC group had a mean HbO level of 1.73 μM (SD = 0.85 μM), whereas the aMCI group had a mean HbO level of 1.22 μM (SD = 0.80 μM). At span level 9, the NC group had a mean HbO level of 1.81 μM (SD = 0.91 μM), whereas the aMCI group had a mean HbO level of 1.25 μM (SD = 0.80 μM). In addition, the naMCI group exhibited a mean HbO level of 1.75 μM (SD = 0.66 μM) at span level 8 and 1.85 μM (SD = 0.65 μM) at span level 9, which did not differ significantly from the NC group (span level 8: P = 0.96, d = 0.016; span level 9: P = 0.86, d = 0.056). Finally, the differences between the aMCI group and the naMCI group were not significant at span level 8, P = 0.089, d = 0.72, and span level 9, P = 0.061, d = 0.84.

FIGURE 3.

A, HbO of visual memory span task in individuals with NC, aMCI, and naMCI. The error bars represent ±  one standard error of the mean. The asterisks represent the significance levels of group differences in HbO. * P < 0.05, with the NC group showing higher HbO than the aMCI group at span level 8 and span level 9. B, Prefrontal activation maps showing the HbO concentration (in μM) in span level 8 and span level 9 of the visual memory span task in NC, aMCI, and naMCI groups. Data from individual channels were used in creating all activation maps. Red color represents greater activation, whereas blue color represents less activation. C, ROC curve of the HbO in span level 9 for detecting individuals with aMCI from NC. aMCI, amnestic mild cognitive impairment; AUC,  area under the ROC curve; HbO, oxyhemoglobin; naMCI, non‐amnestic mild cognitive impairment; NC, normal cognition; ROC, receiver operating characteristic.

3.1.4. Discriminant analysis between NC and MCI subgroups

Next, to assess the capabilities of fNIRS in distinguishing among NC, aMCI, and naMCI, a stepwise discrimination analysis was conducted with HbO data at all span levels entered. Two separate analyses were carried out to differentiate between the NC and aMCI groups and naMCI groups.

The stepwise binary logistic regression analysis revealed that HbO at span level 9 was the significant predictor in the prediction model for aMCI, accounting for 8.1% of the variance in group membership (Cox and Snell R²  = 0.081). This finding represented a significant improvement compared to the baseline model without any predictors (χ² (1) = 4.37, P = 0.037). Additionally, goodness of fit was confirmed through the Hosmer and Lemeshow test (χ² (8) = 6.72, P = 0.57). The ROC analysis demonstrated a significant area under the curve (AUC) of 0.67 (P = 0.034, Figure 3C), indicating an accuracy of 69.23%, a sensitivity of 72.97%, and a specificity of 60.00% for a cut‐off value of 1.12 μM. Participants with HbO levels higher than the cut‐off were classified as NC, while those with HbO levels lower than the cut‐off were classified as aMCI. Furthermore, even after incorporating demographic factors such as age, sex, and level of education, only HbO at span 9 was selected as a significant predictor.

In terms of differentiating between the NC group and the naMCI subgroups, the stepwise binary logistic regression analysis did not identify any HbO variables as significant predictors (P = 0.44–0.99).

3.2. Comparison of the fNIRS results among the NC group and different SMC groups

3.2.1. Demographic information and neuropsychological assessment performance

Table 3 presents the demographics and neuropsychological assessment performance of the NC, mild SMC, moderate SMC, and severe SMC groups. There was no significant group difference, P = 0.14 to 0.95, indicating that all the groups were comparable in terms of age, sex, education, and other cognitive abilities.

TABLE 3.

Demographic information and neuropsychological assessment performance of individuals with normal cognitive function, mild SMC, moderate SMC, and severe SMC.

	NC (n = 37)		Mild SMC (n = 24)		Moderate SMC (n = 33)		Severe SMC (n = 29)
Variables	M	SD	M	SD	M	SD	M	SD	F/χ2	p
Demographics
Age (year)	65.39	6.49	65.76	5.31	66.59	7.02	63.26	7.11	1.39	0.25
Sex (F/M)	24/13		16/8		24/9		24/5		2.91	0.41
Education (year)	13.59	3.97	13.04	3.21	13.64	3.34	12.72	3.59	0.47	0.70
Hong Kong List Learning Test
Total learning	29.30	5.08	31.58	4.96	30.27	5.50	29.07	5.42	1.28	0.29
10‐minute delayed recall	11.08	2.47	11.50	2.73	11.21	2.69	10.31	2.84	1.00	0.39
30‐minute delayed recall	11.11	2.46	11.29	2.33	10.85	2.79	10.07	3.10	1.14	0.33
Recognition	88.34	10.74	93.49	0.03	88.26	12.38	88.79	10.87	1.48	0.22
Category Fluency Test
Animal	19.35	4.81	20.00	3.96	19.88	5.73	19.24	4.87	0.17	0.92
Transportation	15.22	3.88	13.39 a	3.75 a	14.27	3.86	13.24	3.60	1.85	0.14
Boston Naming Test
Spontaneous naming	25.62	2.45	25.88	2.72	25.03	2.05	25.52	2.11	0.69	0.56
Shape Trail Test
Trail A completion time (seconds)	42.75	13.00	39.95	8.21	41.46	11.50	43.31	15.58	0.39	0.76
Trail B completion time (seconds)	98.39	26.98	94.44	19.10	106.52	31.63	98.78	38.65	0.83	0.48
Five‐Point Test
Total unique design	28.78	9.20	28.00	7.28	27.73	7.61	28.28	7.36	0.11	0.95
HK‐MoCA	27.51	2.34	27.92	1.61	27.03	2.39	27.28	1.31	0.96	0.41

Abbreviations: NC, normal cognition; HK‐MoCA, Hong Kong version of Montreal Cognitive Assessment; SD, standard deviation; SMC, subjective memory complaints.

^a One missing value.

3.2.2. Behavioral performance of visual memory span task

In terms of their visual memory span score, the NC, mild SMC, moderate SMC, and severe SMC achieved average scores of 5.86 (SD = 1.32), 5.83 (SD = 1.43), 5.70 (SD = 1.31), and 6.17 (SD = 1.34), respectively. As expected, there was no significant difference in performance across these four groups, F(3,119) = 0.67, P = 0.57, ${\eta }_p^2$ = 0.017. Post hoc comparisons also did not reveal group differences between any of the two groups, P = 0.17 to 0.93.

Next, the HbO levels during the visual memory span task were measured by repeated measures ANOVA. The results showed a non‐significant group × span interaction, F(21,325) = 0.89, P = 0.60, ${\eta }_p^2$ = 0.052, but a significant main effect of group, F(3,119) = 4.40, P = 0.006, ${\eta }_p^2$ = 0.10. The result suggested that there was a significant group difference in HbO, but the difference was not specific to span level. Thus, HbO averaged across all span levels, the severe SMC group (M = 0.67 μM, SD = 0.58 μM) showed significantly lower HbO than the NC group (M = 1.07 μM, SD = 0.58 μM, P = 0.005, d = 0.70) and the mild SMC group (M = 1.20 μM, SD = 0.51 μM, P = 0.001, d = 0.97), and a trend of lower HbO than the moderate SMC group (M = 0.95 μM, SD = 0.60 μM, P = 0.059, d = 0.47). In addition, when analyzing the performance at each span, results showed that there were significant differences from span level 4 to span level 9, F(3,cc119) = 2.93 to 4.95, P = 0.003 to 0.037, ${\eta }_p^2$ = 0.069 to 0.11 (Figure 4A,B). Post hoc comparison showed that both the NC and mild SMC group showed a significantly higher HbO level than the severe SMC group from span level 4 to span level 9, P = 0.0007 to 0.033, d = 0.50 to 1.11, and the moderate SMC group showed a significantly higher HbO level than the severe SMC group from span level 7 to span level 9, P = 0.021 to 0.049, d = 0.51 to 0.61.

FIGURE 4.

A, HbO of visual memory span task in individuals with NC, mild SMC, moderate SMC, and severe SMC. The error bars represent ±  one standard error of the mean. The asterisks represent the significance levels of group differences in HbO. The letters represent the results of the post hoc comparison: ^aNC > Severe SMC, ^bMild SMC > Severe SMC, ^cModerate SMC > Severe SMC. B, Prefrontal activation maps showing the HbO concentration (in μM) from span level 4 and span level 9 of the visual memory span task in NC, mild SMC, moderate SMC, and severe SMC groups. Data from individual channels were used in creating all activation maps. Red color represents greater activation, whereas blue color represents less activation. C, ROC curve of the HbO in span level 9 for detecting individuals with severe SMC from NC. ^* P < 0.05, ^** P < 0.01. AUC,  area under the ROC curve; HbO, oxyhemoglobin; NC, normal cognition; ROC, receiver operating characteristic; SMC, subjective memory complaints.

3.2.3. Correlations between hemodynamic response and memory test performance

While there was an association between self‐reported SMC measures and HbO, the HbO was analyzed with their performance on a memory test. Correlation analysis between the HKLLT performance and HbO at all span levels showed that the 30‐minute delayed recall was significantly positively associated with the HbO at span level 8 and span level 9, r = 0.19 to 0.20, P = 0.030 to 0.031, and the total learning score was significantly positively associated with HbO from span level 7 to span level 9, r = 0.18 to 0.20, P = 0.028 to 0.047 (Figure 5). These findings suggest that among normal older adults, their memory ability was associated with HbO levels, with better memory ability showing higher task‐related HbO levels.

FIGURE 5.

Relationship between HbO of visual memory span task from span level 7 to span level 9 and the 30‐minute delayed recall and total learning of the HKLLT. HbO, oxyhemoglobin; HKLLT, Hong Kong List Learning Test.

3.2.4. Discriminant analysis among the NC group and different SMC groups

Next, a discrimination analysis was conducted to assess the capabilities of fNIRS in differentiating between NC, with mild SMC, moderate SMC, and severe SMC. All nine span levels of HbO were included in the stepwise discrimination analysis. The results indicated that only HbO at span level 8 was included in the prediction model, achieving an accuracy of 33.3%.

The unsatisfactory classification accuracy may be attributed to the non‐significant difference in HbO between NC, mild SMC, and moderate SMC groups. Therefore, the classification analysis was performed separately between the NC group and the other SMC subgroups. For the NC and severe SMC group, the stepwise binary logistic regression results demonstrated that HbO at span level 8 was a significant predictor in the prediction model, explaining 14.9% of the variance in group membership (Cox and Snell R²  = 0.15). This finding represented a significant improvement compared to the baseline model that lacked any predictors (χ² (1) = 10.64, P = 0.001). Furthermore, the goodness of fit was confirmed through the Hosmer and Lemeshow test (χ² (7) = 7.51, P = 0.38). The ROC analysis revealed a significant AUC of 0.71 (P = 0.001, Figure 4C), indicating an accuracy of 69.70%, a sensitivity of 72.97%, and a specificity of 65.52% for a cut‐off value of 1.25 μM. Participants with HbO levels higher than the cut‐off were classified as NC, while those with HbO levels lower than the cut‐off were classified as severe SMC. Additionally, even after incorporating demographic factors such as age, sex, and level of education, only HbO at span level 8 was selected as a predictor.

In terms of differentiating between the NC group and the mild and moderate subgroups, the stepwise binary logistic regression analysis did not identify any HbO variables as significant predictors (NC vs. mild SMC: P = 0.32–0.64; NC vs. moderate SMC: P = 0.26–0.84).

4. DISCUSSION

The primary objective of this study was to assess the efficacy of fNIRS in differentiating between individuals with NC and two distinct types of MCI (aMCI and naMCI). In addition, the study sought to compare the activation patterns during task performance between older adults with SMC and those without. The findings indicated that, compared to individuals with NC, those with aMCI exhibited significantly lower levels of HbO during the visual memory span task, and this difference was not observed in individuals with naMCI. Furthermore, the comparison of NC older adults with varying degrees of SMC revealed that those with severe SMC, who reported the highest level of memory concerns, displayed significantly reduced prefrontal activation compared to the NC group. Therefore, the study suggests that fNIRS holds promise as a valuable tool for identifying individuals at the early stages of dementia.

Previous studies have primarily focused on comparing the hemodynamic activation patterns between NC and MCI individuals, with less attention paid to the differences between MCI subtypes. With the existing literature, most studies examined the levels of lateralization rather than the overall activation pattern. ³⁰ , ⁵³ In the present study, it was found that only the aMCI group, and not the naMCI group, showed lower HbO levels than the NC group at span levels 8 and 9 during a visual memory span task. This task paradigm, which is a memory task, ⁴⁷ may be more sensitive in revealing lower activation in individuals with memory impairment, suggesting that the hemodynamic activation patterns of different MCI subgroups may depend on the cognitive domain of the task used. Furthermore, considering that aMCI has a higher risk of progressing to AD, ¹² , ⁵⁴ the most common type of dementia, it holds significant clinical value to understand the predictive potential of HbO in predicting future cognitive decline. Moreover, identifying specific hemodynamic features associated with different subtypes of MCI can provide insights into the underlying mechanisms and may shed light on early detection and intervention strategies.

While it might be expected that differences in brain activation between aMCI and naMCI would correspond to differences in behavioral task performance, it is not uncommon to observe disparities in research among older adults with different levels of cognitive function. ⁵⁵ , ⁵⁶ , ⁵⁷ It has been reported that there can be discrepancies in brain activation without significant differences in behavioral task performance. ⁵⁵ , ⁵⁶ , ⁵⁷ This phenomenon has been observed in studies comparing individuals with NC and MCI and within MCI subgroups, where brain activation differences were detected despite no significant behavioral task differences. ²⁸ , ⁵⁵ , ⁵⁶ , ⁵⁷ These findings suggest that fNIRS may possess sensitivity in detecting abnormalities between aMCI and naMCI, surpassing the sensitivity of traditional experimental task paradigms.

The present study also investigated the relationship between HbO and performance on neuropsychological assessments, specifically memory assessment scores, among normal individuals with varying degrees of SMC. The results indicated a positive correlation between HbO levels during cognitively demanding tasks (e.g., span level 8 and span level 9) and 30‐minute delayed recall and total learning scores on the HKLLT. Higher levels of HbO were associated with better memory performance. These findings align with a previous study, ³⁹ which found that lower fNIRS activation patterns were linked to lower memory test performance in individuals with SMC. Importantly, the present study demonstrated that the positive association between HbO and memory test scores was not limited to individuals with SMC but was also observed when considering the entire non‐MCI population. Because these individuals have normal neuropsychological assessment performance and are categorized as cognitively normal (NC), the positive association between HbO levels and objective memory test scores highlights the potential of using HbO as a biomarker for memory performance, regardless of cognitive status.

Previous studies have reported mixed findings regarding the use of SMC to identify individuals at risk of dementia, potentially due to the heterogeneous nature of the construct. ¹³ Therefore, the present study examined prefrontal hemodynamics among normal individuals with varying levels of SMC during the computerized visual memory span task. Consistent with our hypothesis, the severe SMC group showed lower HbO levels from span level 4 to span level 9 compared to the NC and the mild SMC group, while the severe SMC group showed lower HbO levels than the moderate SMC group from span level 7 to span level 9. A consistent pattern was observed between the aMCI group and the NC group, with the aMCI group showing lower HbO levels than the NC group at span 8 and span 9. The activation patterns observed in the severe SMC group resembled those reported in previous fNIRS studies, with lower activation observed between MCI/dementia and NC, ²⁸ , ⁵⁸ which is believed to indicate a lack of cognitive resources to sustain cognitive effort and meet the cognitive demands of the task. ²⁸ , ⁵⁹ Taken together, these findings suggest that individuals with severe MCI may exhibit abnormal brain activation before objective cognitive impairment becomes apparent. This indicates that SMC may represent a preceding stage of MCI. In addition, the significant difference in HbO levels observed between the NC and severe SMC groups, but not between the NC and mild SMC group, suggests that the mixed findings observed when comparing the NC and SMC groups may be attributed to the lack of considering an individual's level of SMC.

The results show that there are trends indicating lower HbO levels in severe SMC compared to aMCI and higher HbO levels in mild SMC compared to NC. However, it is important to note that the range of HbO for aMCI was ≈ 0.13 to 0.17 μM, while the range for severe SMC was ≈ 0.23 to 1.25 μM. The independent sample t tests did not yield statistically significant results, t = 0.29 to 1.62, P = 0.11 to 0.54, d = 0.09 to 051, for comparing aMCI and severe SMC. Similarly, non‐significant results were found between mild SMC and NC, t = 0.46 to 0.97, P = 0.34 to 0.65, d = 0.12 to 0.25. For the trend between severe SMC and aMCI, severe SMC refers to individuals who endorse subjective memory problems without exhibiting objective memory problems, whereas aMCI demonstrates objective memory impairments with different levels of SMC. This suggests that subjective memory decline and objective memory impairments may interact and influence hemodynamics. On the other hand, the difference between NC and mild SMC may indicate a compensatory mechanism wherein the mild MCI group uses additional available cognitive resources to complete the task. ⁶⁰ However, given that the present study is exploratory, further investigations are warranted to understand these trends better.

Given that the present study did not use other biomarkers, such as PET, to assess the recruited sample, it would be valuable for future research to compare fNIRS to other biomarkers. Moreover, it would be clinically significant to understand the underlying neurobiological mechanisms that fNIRS can capture. Exploring these mechanisms could provide valuable insights into the effectiveness of fNIRS in identifying individuals with MCI and SMC. Conducting a comprehensive comparison and thorough mechanistic exploration would greatly bolster the credibility and effectiveness of fNIRS as a potential non‐invasive biomarker.

In conclusion, the present study revealed reduced prefrontal activation patterns in individuals with aMCI compared to those with NC. This pattern was consistently observed between normal individuals with severe SMC and those without SMC, indicating that fNIRS could be considered a potential biomarker for individuals with SMC and MCI. Identifying individuals in the early stages of dementia is crucial for early intervention. Therefore, further investigation into the role of fNIRS measures in predicting future cognitive decline, as well as the development of MCI and dementia, is warranted.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

The study was performed in accordance with the Declaration of Helsinki and has been approved by the Joint Chinese University of Hong Kong – New Territories East Cluster Clinical Research Ethics Committee (CREC Ref. No.: 2022.633). All the participants have provided informed consent before the experiment.

Supporting information

Supporting Information

Lee T‐L, Guo L, Chan AS. fNIRS as a biomarker for individuals with subjective memory complaints and MCI. Alzheimer's Dement. 2024;20:5170–5182. 10.1002/alz.13897