Alteration of glucose neurometabolism and brain morphology in mild behavioral impairment: a neuroimaging study on cognitively healthy individuals

Abstract

INTRODUCTION

Regional brain enlargement and increased glucose neurometabolism do not normally indicate neurodegeneration. Mild behavioral impairment (MBI) occurring in cognitively healthy individuals, however, might be significantly affected by these processes.

METHODS

Three hundred forty‐one cognitively normal individuals were analyzed using partial least squares (PLS) regression to determine neurometabolic and brain volumetric determinants of MBI scores. Radiolabeled glucose non‐displaceable binding potential and volumes of Schaefer homooxygenation parcels were screened as regressors. PLS models were evaluated with cross‐validation, scrambling, and bootstrapping.

RESULTS

Hypermetabolism in the left entorhinal cortex and hypertrophy in the left dorsal attention network B contribute substantially to the MBI total score. Significant morpho‐functional interactions were observed between the right middle temporal cortex and the right salience‐ventral attention B network.

DISCUSSION

MBI might depend on cerebral bioenergetic processes and morpho‐functional interactions. Relative hypermetabolism and hypertrophy could be considered specific biomarkers of MBI indicating a prodromal maladaptive neural response.

Highlights

• Glucose hypermetabolism contributes to MBI.

• The volume of homooxygenation brain parcels modulates the neurometabolic effects.

• Neurometabolism is varyingly associated with brain parcel volumes.

1. BACKGROUND

Mild behavioral impairment (MBI) is a known prodrome of cognitive decline that is mostly assessed through questionnaire‐based severity scores within five domains of persistent psychological and neuropsychiatric symptoms (NPSs) (i.e., apathy, anxiety, impulsivity, disinhibition, and psychosis, as given by the MBI checklist [MBI‐C]). ¹ , ² Despite its lower specificity, a growing body of evidence from epidemiological studies confirms that MBI has a prevalence of 17% among cognitively healthy individuals, while prevalence rises to 35.8% in those with subjective cognitive decline and 45.5% in individuals with mild cognitive impairment (MCI). ³ These statistically significant differences align with findings from another study in which 59% of cognitively healthy participants reported at least one NPS prior to the onset of MCI. Notably, 30% of this MCI subgroup later progressed to Alzheimer's disease (AD). ⁴

Thanks to considerable advances in recent neurodiagnostic techniques, numerous research efforts have been focused on identifying neuroimaging and biochemical correlates of MBI scores as indicators of elevated dementia risk. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ MBI has also been associated with blood‐based biomarkers such as the amyloid beta 42/40 ratio, phosphorylated tau, neurofilament light chain, glial fibrillary acidic protein, chitinase‐3‐like protein 1, hippocampal and entorhinal cortex atrophy, decreased connectivity in the default mode, salience, and frontoparietal networks, and variants in the brain‐derived neurotrophic factor or apolipoprotein E genes (APOE). ⁹ , ¹⁰ , ¹¹ , ¹² Unfortunately, the mechanism of MBI and its metabolic aspects remain poorly understood.

Accordingly, a recent study on 36 individuals without dementia failed to detect any significant association between sole glucose metabolism and the MBI‐C total score. ¹³ Other studies reported on a statistically significant linear relationship between the average glucose hypometabolism within an ad hoc set of brain parts (e.g., posterior cingulate cortex and temporal gyrus) and the MBI total score for individuals with MCI. ¹⁴ , ¹⁵ Notably, the hypothesis that glucose hypometabolism drives MBI has been challenged by findings of glucose hypermetabolism in certain brain regions, including the precuneus, prefrontal cortex, and posterior cingulate cortex, which may represent a key stimulus for the emergence of MBI symptoms. ¹⁶ , ¹⁷ , ¹⁸ These conflicting results highlight an unresolved knowledge gap regarding the relationship between cerebral glucose metabolism and MBI, underscoring the need for further investigation in the context of neurodegeneration.

In this study, special attention was directed to exploring glucose neurometabolism from [¹⁸F]fluorodeoxyglucose positron emission tomography (FDG‐PET) and volume changes of Schaefer connectively homogeneous brain parcels from T1‐weighted magnetic resonance imaging (MRI) scans (T1‐MRI) in individuals with MBI who remain cognitively healthy. In addition, statistical interdependence between metabolic and structural brain descriptors was evaluated to derive a specific morpho‐functional signature of MBI in cognitively healthy individuals.

RESEARCH IN CONTEXT

1. Systematic review: Online databases (e.g., PubMed) confirm that existing neuroimaging studies typically examine neurometabolic or morphological correlates of MBI separately. Although MBI scores are linked to both regional brain atrophy and glucose hypometabolism, integrated analyses of morpho‐functional interactions associated with MBI remain largely unexplored.

2. Interpretation: This study uses glucose binding potentials and homooxygenation brain parcel volumes to uncover novel aspects of MBI. The findings support the hypothesis that patterns of hypermetabolism and hypertrophy are relevant to MBI. The uniqueness of the study consists in its detection of links between neurometabolism and oxygenation connectome units in the brain.

3. Future directions: MBI is influenced by distributed morpho‐functional interactions across multiple brain systems rather than by a single localized pathology. Future work could extend these findings through (a) longitudinal analyses of neurometabolism and structural brain changes, (b) detailed analyses of neurometabolism and attentional networks, and (c) investigation of neurometabolism and sensomotoric networks.

2. METHODS

2.1. Involved participants

This study involved 341 participants (174 women) retrospectively selected from the Alzheimer's Disease Neuroimaging Initiative (ADNI, Principal Investigator: Michael W. Weiner, MD) (for further details see: https://adni.loni.usc.edu/). The main selection criteria for this study were set to include any individuals present in ADNI who had complete raw magnetization prepared rapid acquisition with gradient echo (MPRAGE) T1‐weighted MRI scans (T1‐MRI), raw baseline FDG‐PET examination with six emission scans over 30 min, and Neuropsychiatric Inventory Questionaire (NPI‐Q) for MBI scores. Secondary inclusion criteria restricted the analyses only for subjects with normal cognition whose total MBI score was found to be stable within a sliding time window of 6 months and where the acquisition time difference with respect to both T1‐MRI and FDG‐PET experiments did not exceed 1 year. A flowchart showing the selection process of neuroimages is given in Figure 1.

FIGURE 1.

Flowchart of steps to preselect 341 individuals with complete T1‐MRI, FDG‐PET, and NPI‐Q data from the ADNI database.

2.2. Ethics

ADNI representatives guaranteed adherence to the bioethical principles according to the Helsinki Declaration.

2.3. Examined variables

2.3.1. Cognitive diagnostic criteria

This study adopted the main cognitive diagnostic criteria of ADNI to classify subject cognitive status into cognitively normal (CN), MCI, and AD categories based on Clinical Dementia Rating, Mini‐Mental State Examination (MMSE), and Wechsler Logical Memory II subscale testing adjusted for years of education. ¹⁹ , ²⁰ , ²¹ For all analyses, only the CN group (n = 341) was selected.

2.3.2. Neurobehavioral evaluation

The main operationalization of MBI in this work was inferred from the NPI‐Q data available in ADNI utilizing simple algebraic calculations reported by Ismail et al. ¹ , ²² , ²³ To conform to the International Society to Advance Alzheimer's Research and Treatment–Alzheimer's Association criteria for MBI, the five MBI domains were approximated as sums of NPI‐Q subscore severities according to the published rules. ²⁴ MBI total score was defined as the sum of all five MBI subscores. MBI status/syndrome was defined as positive (MBI+) when MBI total score was greater than 1 and negative (MBI−) if equal to 1 or 0.

2.3.3. Additional variables

To further evaluate genetic and exposome associations with MBI, a set of various neurocognitive scores, neuropsychiatric scores, sociodemographic characteristics, biomarkers, clinical data, and pharmacotherapeutic profiles available in the ADNI database was statistically analyzed. Disclosing associations of such variables with MBI syndrome in the studied cohort may account for factors that cannot easily be derived with neuroimaging, and they thus can help elucidate a fuller complexity of MBI.

2.4. Processing neuroimages

2.4.1. Image acquisition

T1‐MRI scans, downloaded from ADNI in 2025, were recorded as MPRAGE sequences by various scanners at different resolutions and two magnetic field strengths (1.5T, 3.0T). The basic parameters of T1‐MRI are represented with time repetition of 1.0 to 2.3 s, time echo of 0.28 to 0.41 s, time inversion of 0.9 s, 3D volume dimensions of 128 to 256 × 192 to 256 × 170 to 256 and voxel sizes of 0.9 to 1.3 mm. This set of T1‐MRI neuroimages involved only raw data in DICOM format that were fully processed within this study.

Similarly, FDG‐PET neuroimages, acquired using different scanning systems at various settings, were downloaded from the ADNI database in 2025 as raw files in DICOM, MRRT, and ECAT formats and originally processed in this study. All FDG‐PET raw images had attenuation correction and comprised six emission scans recorded after 5 min over 30 min. The dimensions of FDG‐PET images were 128 to 400 × 128 to 400 × 47 to 207 with voxel sizes of 0.2 to 3.3 mm. The FDG‐PET studies were performed according to ADNI standard protocol and represent basal cerebral FDG metabolism at resting state.

2.4.2. Analyses of T1‐MRI images

The complete set of T1‐MRI scans was processed with FreeSurfer 8 in a high‐performance computer cluster, applying standard functions for comprehensive morphometric neuroimage examination, which involved normalization, automatic correction, morphing to the standard MNI152 space, main brain tissue segmentation, registration, parcellation, and labeling, surface‐based and volumetric statistical analyses of cortical and subcortical brain regions. ²⁵ Morphing and segmentation of T1‐MRI scans were performed using updated convolutional neural networks tools integrated in FreeSurfer 8.

Further analyses and image quality ranking (IQR, 84.7 ± 1.2%) of T1‐MRI scans were performed in CAT12 under MATLAB 2024b, using default settings for complex structural brain description. ²⁶ Using these computational methods, 341 T1‐MRI scans were thoroughly checked and correctly processed to provide the necessary structural files to analyze FDG‐PET images as well as 200 geometrical brain parameters and the total volume of T1‐based white matter hypointensities for statistical studies of MBI and neuroanatomic correlates. The geometrical parameters (e.g., volumes of gray and white matter) were determined with the Schaefer 17–100 neuroatlas in CAT12, which represents functionally homogeneous brain regions with similar oxygenation connectomics according to a gradient‐weighted Markov random field approach. ²⁷

2.4.3. Analyses of FDG‐PET images

All FDG‐PET images were analyzed in FreeSurfer 8 immediately following the structural analyses of T1‐MRI using shell scripts for distributed and parallelized calculations in a high‐performance computer cluster. The processing algorithm involved motion correction, creation of a mean template of six consecutive emission FDG‐PET scans, registration of the mean template with T1‐MRI main brain tissue segments, point spread function (PSF) reconstruction and partial volume error correction (PVC) with the symmetric geometric transfer matrix (GTM) and Muller‐Gartner threshold of 0.01, and correction of tissue fraction effect.

To determine the kinetic parameters of FDG uptake, a multilinear reference tissue model with two parameters (MRTM2) was implemented in the processing pipeline using the cerebellar cortex as the reference region and caudate, putamen, and pallidum as high‐binding regions. ²⁸ MRTM2 afforded in particular the non‐displaceable binding potential (BP_ND) of FDG, its relative clearance rate constant from the reference brain region, and relative apparent clearance rate constant from the brain regions of interest. The kinetic MRTM2 modeling provided average kinetic quantity estimates for 97 cerebral regions of interest, which were statistically analyzed in further steps. Only BP_ND values were utilized in the final PLS models as they provided the statistically best criteria in preliminary tests.

2.5. Statistical analyses

2.5.1. Descriptive statistics

Besides the NPI‐Q‐related MBI scores, the studied cohort of 341 individuals was comprehensively described with a set of neurocognitive parameters, depression scores (i.e., Geriatric Depression Score [GDS]), practical ability scores (i.e., Functional Activities Questionnaire [FAQ]), sociodemographic characteristics, selected genetic properties, and clinical biomarkers. Additionally, NPI total scores were analyzed as well to check their correspondence with NPI‐Q data, which were used for MBI score estimation in this study.

The aforementioned numerical and categorical properties were statistically analyzed after stratification by MBI status. Normally distributed numerical data proved by Shapiro‐Wilk test were described by arithmetic mean and sample standard deviation, for differently distributed data, median and interquartile range were applied. Categorical data were characterized by the category frequencies and percent proportions.

For the evaluation of statistical differences of the monitored numerical characteristics with a normal distribution, independent two‐sample t‐test and Cohen's d for effect size were applied. Data exhibiting a non‐normal distribution were evaluated with a Mann–Whitney U test, while categorical data were analyzed with a chi‐squared test of independence and φ effect size. All descriptive statistical analyses were performed in MATLAB 2024b.

2.5.2. Statistical modeling of MBI

The first aim of this study was to develop a statistical model of MBI total score utilizing neurometabolic quantities from FDG‐PET (i.e., BP_ND) and morphological descriptors from T1‐MRI as independent variables.

The statistical models were built using a PLS method with an integrated algorithm for the removal of uninformative variables in MATLAB 2024b. This pruning technique is based on stepwise backward elimination of regressors that do not contribute to the overall predictive power of the PLS model within leave‐one‐out (LOO) cross‐validation and exhibit unstable or vanishing behavior of β coefficients. ²⁹ Since the instability of β coefficients, estimated by standard deviation reaching its maximum value in the complete set of regressors, and their relatively minimal mean absolute value indicate irrelevant information content, they were iteratively eliminated from the set of independent variables until the final PLS model reached maximal predictivity and robustness. This enhanced PLS method was used to screen the completely normalized matrix of neurometabolic region‐based FDG indices and morphometric variables to find the best PLS regressors for the MBI total score (i.e., single dependent variable). The main criterion for discerning the optimal PLS models was the highest global LOO cross‐validated coefficient Q ² _LOO in the set of 10 PLS models with one to 10 latent variables (LVs).

2.5.3. Morpho‐functional interactions in MBI

A second aim of this study was to elucidate the relationships between the cerebral FDG metabolism and changes in the Schaefer parcel volumes in the brain with respect to MBI. Based on our previous studies, FDG metabolism indexes and brain structural parameters must be involved in statistical modeling to account for persistent behavioral changes, but their mutual statistical interaction is not clear. ³⁰

Morpho‐functional interactions were evaluated by splitting the group of regressors selected in the PLS model of the MBI total scores into functional (i.e., FDG neurometabolic indices) and structural (i.e., Schaefer brain parcel volumes) parts. Then, the FDG neurometabolic regressors were used as multiple dependent variables in a new PLS modeling to find the best fit with the second set of structural brain parameters. These PLS models were evaluated with three to 10 latent variables, and the model achieving the highest stability in LOO cross‐validation (i.e., maximal Q ² _LOO ) was bootstrapped in 10,000 repetitions with replacement to estimate mean β coefficients, their 95% confidence intervals (CIs), probabilities, and variable importance in projection. The morpho‐functional interactions are therefore related to MBI total score, and their effects are expressed with statistically significant β coefficients (i.e., p < 0.05).

2.5.4. Validation studies

The PLS models were considered valid if the Q ² _LOO values were at least 20% higher than those resulting from Y‐scrambling (i.e., ^sc Q ² _LOO values that are required to be negative). In addition, the probabilities of Q ² _LOO were estimated using a Fisher‐Snedecor F statistical distribution.

With respect to the validation of individual PLS β coefficients, bootstrap analyses of the complete PLS model using 10,000 repetitions with a replacement technique were carried out to derive the 95% CIs of β coefficients, their statistical significances, and variable importance in projection (VIP). All validation studies were conducted in MATLAB 2024b.

3. RESULTS

3.1. Basic cohort statistics

An overview of the basic statistical characteristics of the sample, involving sociodemographic properties, genetic and clinical data, and their significances is given in Table 1. The most important sociodemographic factor associated with MBI syndrome (MBI+) is marital status different than married (p = 0.04, φ = 0.11).

TABLE 1.

Descriptive statistical analysis of sociodemographic, neurocognitive, genetic, clinical, and neurobehavioral characteristics of the involved cognitively healthy participants (n = 341), stratified by MBI status.

Descriptor	MBI−	MBI+	p	d/φ
Number	300	41	—	—
Age (years)	75.5 ± 6.3	76.9 ± 8.1	0.21	0.19
Women	154 (51%)	20 (49%)	0.76	0.02
Education (years)	16.7 ± 2.6	15.9 ± 2.3	0.07	0.33
Languages	1.0 ± 0.2	1.0 ± 0.0	0.43	0.00
Handedness (right)	274 (91%)	35 (85%)	0.23	0.07
Marital status (married)	221 (74%)	27 (65%)	0.04	0.11
Ethnic (non‐Hispanic)	279 (93%)	39 (95%)	0.84	0.01
Race (White)	265 (88%)	36 (87%)	0.74	0.02
MoCA	25.5 ± 2.7	24.7 ± 3.6	0.03	0.25
MMSE	29.1 ± 1.2	28.6 ± 1.2	0.01	0.42
ADAS‐Cog13	8.8 ± 4.4	8.7 ± 5.1	0.92	0.02
FAQ	0.1 ± 0.5	1.3 ± 2.7	<0.01	0.62
GDS	0.8 ± 1.1	1.2 ± 1.9	0.03	0.26
APOE ε2/ε2	2 (1%)	0 (0%)	0.60	0.03
APOE ε3/ε3	171 (57%)	22 (54%)	0.69	0.02
APOE ε4/ε4	6 (2%)	2 (5%)	0.25	0.06
APOE ε2/ε3	43 (14%)	7 (17%)	0.64	0.03
APOE ε2/ε4	7 (2%)	1 (2%)	0.97	0.00
APOE ε3/ε4	71 (24%)	9 (22%)	0.81	0.01
Systolic blood pressure	134.3 ± 16.4	133.1 ± 16.6	0.66	0.07
Diastolic blood pressure	73.4 ± 9.2	72.9 ± 9.2	0.74	0.05
Weight	76.9 ± 15.9	78.8 ± 15.9	0.50	0.12
BMI	27.2 ± 5.0	27.5 ± 4.1	0.81	0.07
Smoking	73 (24%)	11 (27%)	0.73	0.02
Hypertension	147 (49%)	19 (46%)	0.75	0.02
Diabetes	12 (4%)	5 (12%)	0.02	0.13
Heart disease	6 (2%)	2 (5%)	0.25	0.06
Stroke	10 (2%)	2 (5%)	0.61	0.03
Antihypertensives	56 (19%)	8 (20%)	0.90	0.01
Anxiolytics	13 (4%)	2 (5%)	0.87	0.01
Antidepressants	68 (22%)	16 (39%)	0.02	0.19
NPI total	0.7 ± 1.9	9.4 ± 9.0	<0.01	1.34
NPI‐Q total	0.2 ± 0.5	4.3 ± 3.2	<0.01	1.79
MBI total	0.1 ± 0.3	3.7 ± 2.7	<0.01	1.87
Motivation	0.0 ± 0.1	0.4 ± 0.7	<0.01	0.80
Emotional dysregulation	0.1 ± 0.2	1.4 ± 1.3	<0.01	1.40
Impulse dyscontrol	0.1 ± 0.2	1.7 ± 1.6	<0.01	1.40
Social inappropriateness	0.0 ± 0.0	0.2 ± 0.5	<0.01	0.57
Abnormal thoughts	0.0 ± 0.0	0.0 ± 0.2	0.01	0.00

Note: p values given in bold represent statistically significant differences (α < 0.05) between the descriptors estimated for the MBI− and MBI+ groups.

Abbreviations: ADAS‐Cog13, Alzheimer's Disease Assessment Scale–Cognitive Subscale 13; APOE1, apolipoprotein E allele 1; APOE2, apolipoprotein E allele 2; BMI, body mass index; d/φ, effect size given as Cohen's d (numeric data) or φ coefficient (categorical data); FAQ, Functional Activities Questionnaire; GDS, Geriatric Depression Scale; MBI, mild behavioral impairment; MMSE, Mini‐Mental State Examination; MoCA, Montreal Cognitive Assessment; NPI, Neuropsychiatric Inventory; NPI‐Q, Neuropsychiatric Inventory Questionnaire; Number, numbers of involved individuals; p, probability of two‐sample Student's t‐test (numeric data) or chi‐squared test of independence (categorical data).

Lower MMSE and Montreal Cognitive Assessment (MoCA) scores are both associated with MBI syndrome (p = 0.01, d = 0.42; p < 0.01, d = 0.39). Similarly, increased FAQ and GDS scores are significantly elevated with MBI syndrome (p < 0.01, φ = 0.62; p = 0.03, d = 0.26).

APOE genotype is not associated with MBI syndrome. Increased application of antidepressants (p = 0.02, φ = 0.19) and higher frequency of type 2 diabetes mellitus (p = 0.02, φ = 0.13) were found to be significant for the MBI+ subgroup.

Neurobehavioral measures (e.g., Neuropsychiatric Inventory total score, NPI‐Q total score, MBI total score, MBI domain subscores) were significantly elevated in MBI+, with the exception of abnormal thoughts due to their rare occurrence in the studied cohort.

3.2. Statistical model of MBI

The most significant PLS model for MBI total score with maximal R ²  = 0.56, Q ² _LOO  = 0.29 and ^scQ ² _LOO  = −0.24 resulted when seven LVs were employed, representing 73 top predictors (i.e., 16× FDG BP_ND, 55× Schaefer parcel volumes, body weight, and sex). The relationship between the predicted and real MBI total scores, the histogram of residuals, and Q‐Q plot of predictive residuals are illustrated in Figure 2.

FIGURE 2.

(A) Relationship between real and predicted MBI total scores for cognitively healthy individuals (n = 341). Histogram (B) and Q‐Q plot (C) of predictive residuals of MBI scores show considerable distribution normality, which was confirmed with Shapiro‐Wilk test (p = 0.115).

Based on bootstrapping with 10,000 repetitions (with replacement), only FDG BP_ND indices, Schaefer volumes, and confounders exhibiting significant β coefficients (p < 0.05) and mean VIP > 1 were selected for further interpretation (Table 2). For CN, relative FDG hypermetabolism in the left entorhinal (β = 0.424, p = 0.008, VIP = 2.248) and the left middle temporal (β = 0.269, p = 0.003, VIP = 1.064) cortices were associated with increased MBI total score. Hypometabolism contributing significantly to the MBI total score was detected in the right pericalcarine (β = −0.364, p = 0.001, VIP = 1.158), right frontal pole (β = −0.253, p = 0.009, VIP = 1.248) cortices and the left thalamus (β = −0.338, p = 0.013, VIP = 1.423). In the morphological part of the PLS model for MBI total score, seven brain parcels with gray matter volume and five parcels with white matter volume were found to be significantly contributing to model performance. With respect to the best probabilities of β coefficients, greater gray matter volume in the left dorsal attention network B with frontal eye fields 1 (β = 0.518, p < 0.001, VIP = 1.558), atrophy of white matter volume in the left central somatomotor B network 1 (β = −0.422, p = 0.001, VIP = 1.248), and the right salience‐ventral attention B network 1 in the lateral prefrontal part (β = −0.490, p = 0.001, VIP = 1.830) showed the strongest associations with MBI total score. Importantly, the statistical significance of the selected morpho‐functional variables was stabilized when the female sex indicator variable was involved in the set of regressors (β = 0.215, p = 0.028, VIP = 1.374).

TABLE 2.

Overview of the most significant neurometabolic and morphological parameters (p < 0.05 and VIP > 1.0) selected in PLS model of MBI total score in the cognitively healthy group (n = 341, LV = 7, R²  = 0.56, F = 57.95, ‐log(P_F ) > 10, Q² _LOO  = 0.29, ^scQ² _LOO = −0.24).

Brain parcel name	β [95% CI]	p	VIP
B_ctx‐rh‐pericalcarine	−0.363 [−0.584, −0.142]	0.001	1.528
B_ctx‐rh‐middletemporal	0.269 [0.093, 0.444]	0.003	1.064
B_ctx‐lh‐entorhinal	0.424 [0.113, 0.734]	0.008	2.478
B_ctx‐rh‐frontalpole	−0.253 [−0.442, −0.064]	0.009	1.248
B_lh‐thalamus	−0.338 [−0.603, −0.072]	0.013	1.423
G_lDorsAttnB_FEF_1	0.518 [0.24, 0.795]	<0.000	1.558
W_lSomMotB_Cent_1	−0.422 [−0.661, −0.183]	0.001	1.248
W_rSalVentAttnB_PFCl_1	−0.490 [−0.781, −0.199]	0.001	1.830
G_rDefaultA_pCunPCC_1	−0.364 [−0.603, −0.125]	0.003	1.021
G_rDefaultB_PFCv_1	0.324 [0.098, 0.55]	0.005	1.272
W_rVisPeri_StriCal_1	0.323 [0.099, 0.548]	0.005	1.014
W_rDefaultB_PFCv_2	0.517 [0.159, 0.875]	0.005	1.617
G_lDefaultB_PFCl_1	−0.355 [−0.612, −0.098]	0.007	1.235
G_lContA_IPS_1	0.411 [0.09, 0.733]	0.012	1.244
G_rSalVentAttnA_ParMed_1	−0.355 [−0.647, −0.063]	0.017	1.174
G_rContA_PFCl_1	0.314 [0.05, 0.578]	0.020	1.083
Female sex	0.215 [0.023, 0.408]	0.028	1.374
W_lSalVentAttnA_Ins_1	−0.224 [−0.438, −0.01]	0.041	1.028

Note: The full PLS model contains 73 variables. Only the aforementioned set of 18 parameters proved stability in bootstrapping (p < 0.05 and VIP > 1.0).

Abbreviations: B, non‐displaceable binding potential of FDG; β, beta coefficient from PLS; CI, confidence intervals of β determined in bootstrapping with 10,000 repetitions; CN, cognitively normal; ctx, cerebral cortex; F, Fisher‐Snedecor F test; G, gray matter volume of Schaefer brain parcel; LV, number of latent variables; n, number of individuals; p, probability of β coefficients determined in bootstrapping with 10,000 repetitions; Q ² _LOO , coefficient of determination cross‐validated with leave‐one‐out technique; R² , coefficient of determination; ^scQ² _LOO , scrambled coefficient of determination cross‐validated with leave‐one‐out technique in 1000 repetitions; VIP, mean variable importance in projection determined in bootstrapping with 10,000 repetitions; W, white matter volume of Schaefer brain parcel.

To distinguish metabolic and volumetric contributions in the complete PLS model for the CN cohort, the descriptors were separated and summed as products of the normalized FDG BP_ND values or Schaefer parcel volumes, respectively, and the corresponding β coefficients (Figure 3).

FIGURE 3.

Differences in metabolic (A) and volumetric (B) functions in PLS model for studied MBI subgroups (n = 341) (with [+] and without [−] MBI syndrome). (C) Predicted MBI total scores in studied MBI subgroups by full PLS models with complete set of predictors. The numbers above the boxplots denote the statistical significance of the two‐sample t‐tests.

3.3. Morpho‐functional dynamics in MBI

The most important positive interaction occurred between FDG BP_ND in the right middle temporal cortex and the left gray matter volume of dorsal attention network B with frontal eye fields 1 (β = 0.220, p = 0.029, VIP = 1.191), and the second negative interaction was confirmed between BP_ND in the right pericalcarine cortex and the right white matter volume of salience‐ventral attention B network 1 in the lateral prefrontal brain region (β = −0.173, p = 0.035, VIP = 1.348). Other morpho‐functional interactions did not exhibit p < 0.05 within the bootstrapping test of the PLS model. Significant morpho‐functional interactions are displayed in Figure 4.

FIGURE 4.

Significant morpho‐functional interactions (i.e., MBI cerebral signatures) between local neurometabolism (FDG BPND) and volumes of Schaefer homooxygenation brain parcels identified in PLS models of MBI total score in cognitively healthy individuals (n = 341). The numbers close to the connection lines represent significant (p < 0.05) β coefficients derived by bootstrapped multivariate PLS analyses of regressors in MBI statistical models.

4. DISCUSSION

4.1. General association with MBI

In this study, MBI syndrome showed a significant association with unmarried status, cognitive decline (i.e., with respect to MMSE and MoCA), functional activities (i.e., FAQ), administration of antidepressants, and type 2 diabetes mellitus. These findings are consistent with previous studies and confirm the internal consistency of the studied dataset. ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ Importantly, the hypothesis that insulin resistance constitutes a biological mechanism underlying MBI has not yet been proven, although impaired cerebral glucose metabolism clearly disrupts bioenergetic homeostasis, likely contributing to behavioral changes. ³⁷

4.2. Statistical models of MBI

Splitting the derived PLS model of the MBI total score clearly showed that the metabolic function of hypometabolic and hypermetabolic contributors could predict MBI status (Figure 3). Separated volumetric parts of the PLS model showed an even stronger association of the structural changes of Schaefer parcels in the prediction of MBI status. Importantly, the optimal PLS‐based prediction of MBI total scores demonstrated that the combination and interaction of both metabolic and volumetric functions along with the confounders in the full PLS models with 73 variables undoubtedly provided the best MBI status estimate. This evidence proves the efficiency and superiority of the proposed statistical modeling based on morpho‐functional mediation in MBI measures.

Hypermetabolism increasing the MBI total score was detected in the right middle temporal and left entorhinal cortices, which are involved in processing social and emotional semantic contexts as well as in the formation and organization of verbal and episodic memories, respectively. Importantly, the left entorhinal cortex is one of the most reliable neural sites of AD pathology, where the earliest glucose hypometabolism, hyperphosphorylated tau pathology, atrophy, damaged connectivity with the hippocampus, posterior cingulate, and medial prefrontal cortices occur. ³⁸ In contrast, glucose hypermetabolism in the left entorhinal cortex exhibited the highest VIP, 2478, in the developed PLS model, which might be associated with the engagement of high cognitive reserve and increased influence of this region on MBI regulation. Other neurometabolic regressors were related to a significant effect of hypometabolism.

Within the morphological part of the PLS model, a complex set of gray and white matter volumes of Schaefer parcels across the entire brain exhibited statistically significant contributions to the predicted MBI total score. The MBI total score increases when the white matter volume in the right salience‐ventral attention B network in the lateral prefrontal cortex region 1 decreases and the gray matter volume in the left dorsal attention B network with frontal eye field 1 increases. These networks are involved in orienting attention to behaviorally relevant or unexpected stimuli and in selecting and maintaining focus on task‐relevant spatial information. ³⁹

The involvement of atrophy of white matter in the right salience‐ventral attention B network in the regressor set is justifiable as a similar type of neurodegeneration was associated with decreased selective attention scoring in the Flanker task. ⁴⁰ Moreover, the white matter volume of the right salience‐ventral attention B network was found to be negatively related to glucose BP_ND in the right pericalcarine cortex in our morpho‐functional PLS analysis, indicating that shrinkage of the network was associated with increased FDG uptake in the right pericalcarine cortex. Such an non‐trivial relationship suggests a potential compensatory sensory‐attention regulation involved in MBI manifestation in the CN group. This hypothesis on lateralized and disproportionate morpho‐functional pairing in the right hemisphere could be plausible for behavioral changes depending on altered attention or sensory reliance. ⁴¹

Interestingly, the detected effect of structural enlargement of the gray matter volume in the left dorsal attention B network with frontal eye field 1 has not been described in the literature, but functional hyper‐engagement of this network has been observed and might be associated with early neurodegeneration and maladaptive processes in attentional control. ⁴² Based on the nodal stress hypothesis, this kind of hyperactivity can be associated with an induced neuroplasticity of the network region, but it probably increases its vulnerability and precedes atrophy. ⁴³ , ⁴⁴

Morpho‐functional analysis confirmed the statistically significant positive association of the gray matter volume of the left dorsal attention B network with frontal eye field 1 and glucose BP_ND in the right middle temporal cortex (Figure 4). This presupposed cross‐hemispheric link might underscore the potential neurobiological relationship of the left frontal eye field 1 reorganization with increased FDG uptake in the right middle temporal cortex to process perceptual overstimulation. It is likely a maladaptive process affecting different levels of the attention hierarchy. ⁴⁵

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

The ADNI study was approved by each ADNI study site's respective Institutional Review Board, and informed written consent was obtained from all participants.

Supporting information

Supporting Information

Dolezal R. Alteration of glucose neurometabolism and brain morphology in mild behavioral impairment: a neuroimaging study on cognitively healthy individuals. Alzheimer's Dement. 2026;18:e70425. 10.1002/dad2.70425