Skip to main content
NCBI home page
Human Brain Mapping logoLink to Human Brain Mapping
. 2023 Nov 22;45(1):e26529. doi: 10.1002/hbm.26529

Structural, static, and dynamic functional MRI predictors for conversion from mild cognitive impairment to Alzheimer's disease: Inter‐cohort validation of Shanghai Memory Study and ADNI

Zhihan Chen 1,2, Keliang Chen 3, Yuxin Li 1,4, Daoying Geng 1,2,4, Xiantao Li 5, Xiaoniu Liang 3, Huimeng Lu 3, Saineng Ding 3, Zhenxu Xiao 3, Xiaoxi Ma 3, Li Zheng 3, Ding Ding 3, Qianhua Zhao 3,6,7,8,✉, Liqin Yang 1,4,✉; for the Alzheimer's Disease Neuroimaging Initiative
PMCID: PMC10789213  PMID: 37991144

Abstract

Mild cognitive impairment (MCI) is a critical prodromal stage of Alzheimer's disease (AD), and the mechanism underlying the conversion is not fully explored. Construction and inter‐cohort validation of imaging biomarkers for predicting MCI conversion is of great challenge at present, due to lack of longitudinal cohorts and poor reproducibility of various study‐specific imaging indices. We proposed a novel framework for inter‐cohort MCI conversion prediction, involving comparison of structural, static, and dynamic functional brain features from structural magnetic resonance imaging (sMRI) and resting‐state functional MRI (fMRI) between MCI converters (MCI_C) and non‐converters (MCI_NC), and support vector machine for construction of prediction models. A total of 218 MCI patients with 3‐year follow‐up outcome were selected from two independent cohorts: Shanghai Memory Study cohort for internal cross‐validation, and Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort for external validation. In comparison with MCI_NC, MCI_C were mainly characterized by atrophy, regional hyperactivity and inter‐network hypo‐connectivity, and dynamic alterations characterized by regional and connectional instability, involving medial temporal lobe (MTL), posterior parietal cortex (PPC), and occipital cortex. All imaging‐based prediction models achieved an area under the curve (AUC) > 0.7 in both cohorts, with the multi‐modality MRI models as the best with excellent performances of AUC > 0.85. Notably, the combination of static and dynamic fMRI resulted in overall better performance as relative to static or dynamic fMRI solely, supporting the contribution of dynamic features. This inter‐cohort validation study provides a new insight into the mechanisms of MCI conversion involving brain dynamics, and paves a way for clinical use of structural and functional MRI biomarkers in future.

Keywords: Alzheimer's disease, functional connectivity, mild cognitive impairment, resting‐state functional magnetic resonance imaging, support vector machine

Inter‐cohort validation was performed for MRI‐based prediction of MCI conversion. The framework integrated structural, static, and dynamic functional features. Altered dynamic functional features were reported in MCI converters.

graphic file with name HBM-45-e26529-g016.jpg

1. INTRODUCTION

Alzheimer's disease (AD) is an irreversible neurodegenerative continuum involving a lengthy predementia period, and its disease course is characterized by progressive decline of cognitive functions including episodic memory, language facilities, spatial processing ability and executive functions, and accumulation of abnormal proteins (Aβ and tau) in the brain pathologically (Albert et al., 2011; Braak & Braak, 1991; van Oostveen & de Lange, 2021). Mild cognitive impairment (MCI) is the prodromal stage of AD suffering profound cognitive impairment than normal aging but has basically intact ability of daily life. According to previous reports, MCI patients convert to AD at a rate of 10%–15% per year, and more than a half of MCI cases would develop AD longitudinally (Busse et al., 2006; Petersen et al., 2001). Early treatment and interventions in predementia stage may slow the AD process, as suggested by some studies (Molinuevo et al., 2011; Sperling et al., 2011; Wells et al., 2013). These evidences underline the critical role of MCI as a “conversion stage” for disease monitoring and stratified early intervention. Accordingly, it is of great importance to uncover the mechanisms of MCI progression and furtherly explore effective biomarkers for MCI‐to‐AD prediction, which is still an unsolved challenge.

Inter‐cohort validation is a prerequisite for clinical use of biomarkers but always a shortage of neuroimaging‐based studies. In recent years, magnetic resonance imaging techniques, especially their combination with the machine learning methods, are attracting increasing interests for various classification studies in patients with brain disorders including AD continuum (Arbabshirani et al., 2017). Ideally, reliable analysis framework and imaging biomarkers/models should have good reproducibility or generalizability, regardless of the imaging protocols, scanners, and demographics (Rathore et al., 2017). An external validation dataset independent to the one used for biomarker extraction or model training, especially that from other populations with different countries or races/ethnicities, would be an optimal choice.

However, due to the lack of longitudinal datasets involving MCI conversion information, the development and inter‐cohort validation of imaging features or tools are quite difficult for the prediction of MCI to AD. According to a recent review (Ansart et al., 2021), 84.6% of the machine‐learning‐based MCI progression prediction studies used Alzheimer's Disease Neuroimaging Initiative (ADNI) database, a longitudinal dataset of participants recruited in the United States and Canada. Structural magnetic resonance imaging (sMRI, commonly T1‐weighted MRI) was the most popular modality for predicting MCI conversion, and all of these studies were performed on the ADNI dataset (Ansart et al., 2021). Besides, the performance of quantitative sMRI for MCI conversion prediction were largely inconsistent according to previous reports, including a highest prediction accuracy of 93% based on 10 MCI converters, and an accuracy of 65% using 167 MCI converters (Leandrou et al., 2018). Multi‐cohort MRI validation studies concerning AD continuum are very few. For predicting MCI conversion, two multi‐cohort studies using T1 MRI reported accuracy of 64%–75% and AUC of 0.70–0.75 in the external dataset (Bron et al., 2021; Hall et al., 2015), suggesting a certain cross‐cohort effectiveness of sMRI for MCI conversion prediction.

Resting‐state functional MRI (fMRI) is a promising non‐invasive imaging technique for measurements of spontaneous activity and functional architecture of human brain (Biswal et al., 1995; van den Heuvel & Hulshoff Pol, 2010), with a wide application to AD‐ or MCI‐related dysfunction studies (Badhwar et al., 2017). However, in practice, the development of reliable fMRI biomarkers is quite difficult for neurodegenerative disorders (Hohenfeld et al., 2018), especially for MCI conversion problem. First, establishment of an effective fMRI analysis framework is challenging, due to the high sensitivity of fMRI to brain states, the numerous fMRI analysis methods and indices resulting in substantial variation between studies (Hohenfeld et al., 2018; Lee et al., 2013), and the similar clinical and neuropsychological status between MCI converters and non‐converters leading to potentially subtle functional differences (Tan et al., 2014). Second, due to the lack of fMRI data with longitudinal outcome including ADNI, the constructed fMRI predictor would be study‐specific and susceptible to overfitting. For these reasons, there are only several studies utilizing fMRI to predict MCI conversion within 1.5–3 years, based on single dataset with less than 80 MCI converters and non‐converters in total, resulting in barely overlapped fMRI imaging features between studies (Gupta et al., 2020; Hojjati et al., 2017, 2018; Li et al., 2016). Given the limited patients number and the lack of validation in external datasets, the utility of fMRI for MCI prediction still needs more exploration.

Brain function dynamics measured with fMRI is an appealing advance in recent years (Hutchison et al., 2013), including the field of AD and other neurodegenerative disorders (Cordova‐Palomera et al., 2017; Filippi et al., 2019; Jones et al., 2012; Sendi et al., 2021). Unlike traditional static fMRI analysis that are based on the entire time series, dynamic fMRI measurements are conducted after segmenting the time series into short time windows, thus reflecting the variability or stability of brain activity and connectivity overlooked by static fMRI. As reported by two recent studies, dynamic FC outperformed static FC in differentiating AD from controls and other dementia subtype in a multicenter study (Moguilner et al., 2021), and combination of static and dynamical function networks identified early MCI and normal controls with better performance than static networks solely (Kam et al., 2020), suggesting the potential superiority of including dynamic features in fMRI‐based AD studies. Regarding the MCI‐to‐AD conversion problem, whether there is a significant change in brain dynamics before the conversion, and which measurements may predict the conversion, are completely unknown at present.

In this study, we aimed to propose a novel framework by combining sMRI and static and dynamic fMRI and detect effective MRI features for predicting the conversion from MCI to AD within 3 years across different cohorts. In particular, for retrieval of reliable cross‐dataset biomarkers that are also interpretable and convenient for future clinic use, we empirically considered and conducted the following analysis details: (1) Structural and regional functional features were performed based on brain anatomical regions of interest (ROI), instead of sphere ROIs or voxels that may potentially be dataset‐dependent; (2) For static/dynamic fMRI, regional activity and inter‐network FC were both included to characterize two different aspects of brain function; (3) Static and dynamic FC were calculated based on functional components derived from independent component analysis (ICA), which can be leveraged to capture dynamic functional FC properties robust to artifacts and false‐positives (Calhoun & de Lacy, 2017); (4) For external validation cohort, group‐information guided independent component analysis (GIG‐ICA) was utilized for extraction of ICs, assuring correspondence with the internal validation cohort. Two independent cohorts were used: Shanghai Memory Study (SMS) cohort for selection of potentially effective features and internal cross‐validation, and ADNI cohort for these features' external validation, which is different from SMS cohort in population origins, clinical characteristics, scanners, and imaging parameters.

2. MATERIALS AND METHODS

A schematic overview of analysis methods in this study is shown in Figure 1. Details were shown as follows.

FIGURE 1.

FIGURE 1

Open in a new tab

Schematic overview of analysis.

2.1. Participants

2.1.1. Shanghai Memory Study cohort

The Shanghai Memory Study (SMS), established in 2009, is a hospital‐based cohort with participants recruited from the memory clinic of Huashan Hospital, Fudan University, Shanghai, China. SMS is designed to investigate the risk factors, cognitive function, disease progression, and neuroimaging of cognitive disorders with longitudinal follow‐up. This study was approved by the Institutional Review Board (IRB) of Huashan Hospital. Written informed consent was obtained from the participant or the caregiver. The clinical visits and neuroimaging examinations were completed from May 2012 to September 2018.

Diagnosis of MCI fulfilled Petersen's criteria of MCI (Petersen, 2004), and probable and possible AD was diagnosed according to the NINCDS‐ADRDA criteria (McKhann et al., 1984). The APOE genotypes were determined using the TaqMan assay according to the method described previously (Koch et al., 2002). The Mini Mental State Examination (MMSE) (Katzman et al., 1988) was performed for all subjects.

Only participants with both high‐quality whole‐brain T1 and fMRI data at baseline and outcome at 3‐year follow‐up were included. These participants underwent annually clinical and cognitive assessments. Using the diagnostic criteria defined at baseline and at the three‐year time point, we divided patients into two groups, MCI who converted to AD (both possible and probable AD) within 3 years (MCI_C) and MCI who did not convert to AD within 3 years (MCI_NC). Patients with unstable clinical status were excluded, for example, the patient who converted to possible AD at 2‐year follow‐up and reverted to MCI at 3‐year follow‐up. Besides, after the fMRI data preprocessing steps, participants with large head motion were also excluded.

Totally, 41 MCI_C (average age 72 ± 6.7 years; 16 male) and 75 MCI_NC (average age 69 ± 7.8 years; 37 male) were included in this study. Age, years of education, and MMSE score were compared between MCI_C and MCI_NC with two‐sample t‐test. Group difference in gender and APOE E4 carriers were evaluated with Chi‐square test. These statistical analyses were conducted using SPSS statistical software (version 20.0; Chicago, Inc., IL). p value <.05 was considered statistically significant.

2.1.2. ADNI cohort

The ADNI database was publicly available (http://adni.loni.usc.edu). For more details, see www.adni-info.org.

Inclusion and exclusion of MCI‐C and MCI_NC patients were based on the diagnosis variable “DXCHANGE” and “DXCURRENT” in diagnostic summary file of ADNI 1/2/GO subjects: 2 = stable in MCI, 5 = MCI convert to AD, 8 = revision from AD to MCI for “DXCHANGE”; 1 = normal control, 2 = MCI, 3 = AD for “DXCURRENT.” MCI patients with DXCHANGE values of 2 and 5 at 3‐year follow‐up or with DXCURRENT values of stable “2” and from “2” to “3” within 3‐year follow‐up were identified as MCI_NC and MCI_C, respectively. MCI patients with DXCHANGE values of 8 at 1‐year or 2‐year follow‐up and were excluded, which ensured a stable clinical condition during MCI progression within 3 years. The specific flowchart of ADNI cohort data selection was shown in Figure S1.

Accordingly, 25 MCI_C and 77 MCI_NC in ADNI dataset with sMRI and fMRI at baseline were included in this study. The statistical approach of clinical data in ADNI cohort was same with SMS cohort.

2.2. MRI data acquisition

The functional and structural MRI images of SMS cohort were acquired from a Siemens Verio 3 T MRI Scanner. During MRI scanning, all subjects were instructed to keep their eyes closed but not to fall asleep. The rs‐fMRI images were obtained using a single‐shot gradient‐recalled echo planar imaging (EPI) sequence with the following parameters: repetition time (TR) = 2000 ms, echo time (TE) = 35 ms, slice thickness = 4 mm, flip angle = 90°, slices = 33, field of view (FOV) = 256 × 256, matrix size = 64 × 64, number of volumes = 200. High‐resolution 3D sagittal T1‐weighted images were acquired using a magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence: TR = 2300 ms, TE = 2.98 ms, inversion time (TI) = 900 ms, flip angle = 9°, slices = 176, FOV = 240 × 256, matrix size = 1 × 1 × 1.

The MRI images of selected ADNI data were collected using three scanners: Philips 3 T MRI scanner, GE 3 T MRI scanner and Siemens 3 T MRI scanner. The fMRI images were obtained using a EPI sequence with following parameters: A total of 140 volumes (TR = 3000 ms, TE = 30 ms, flip angles = 80°, slice thickness = 3.313 mm, slices = 48) or 197/200 volumes (TR = 3000 ms, TE = 30 ms, flip angles = 90°, slice thickness = 3.4 mm, slices = 48). T1‐weighted images were acquired using a MPRAGE sequence (TR = 6.8 ms, TE = 3.2 ms, flip angle = 9°, slices = 170, FOV = 256 × 256)/(TR = 6.5 ms, TE = 2.9 ms, flip angle = 9°, slices = 211, FOV = 256 × 256)/(TR = 2.3 ms, TE = 3 ms, flip angle = 9°, slices = 176, FOV = 240 × 256) or an IRSPGR sequence (TR = 7.7 ms, TE = 3.1 ms, flip angle = 11°, slices =196, FOV = 256 × 256) with a matrix size = 1 × 1 × 1.

2.3. sMRI features

The FreeSurfer (version 6.0, http://surfer.nmr.mgh.harvard.edu/) (Fischl, 2012) was used to generate structural measures from the T1‐weighted MRI of all subjects. Briefly, the recon‐all processing steps of FreeSurfer includes: removal of non‐brain tissue using a hybrid watershed/surface deformation procedure; automated Talairach transformation; segmentation of the subcortical white matter and deep gray matter volumetric structures; intensity normalization; tessellation of the gray matter and white matter boundary; automated topology correction; and surface deformation following intensity gradients to optimally place the gray/white and gray/cerebrospinal fluid (CSF) borders at the location of the greatest shift in intensity which defines the transition to the other tissue class. In addition to segmentation of 16 subcortical regions of interest (ROI), Desikan–Killiany atlas (Desikan et al., 2006) were used for cortical parcellation, revealing 34 cortical ROI for each hemisphere of each subject. The average cortical thickness of 68 cortical ROIs and gray matter volumes of 16 subcortical ROIs were chosen as candidate sMRI features for further feature selection.

2.4. rs‐fMRI features

2.4.1. Preprocessing

The rs‐fMRI images were preprocessed using DPABI toolbox (http://rfmri.org/dpabi) (Yan et al., 2016). For each subject, the first 10 volumes were removed, and the rest images were corrected for slice‐timing and head motion. The mean frame‐wise displacement (FD) was calculated to represent overall head movement (Power et al., 2014). Participants with a head motion of translation >2.5 mm, rotation >2.5°, or mean FD > 0.2 mm were excluded. Subsequently, the rs‐fMRI images were normalized to the Montreal Neurological Institute (MNI) space using the DARTEL algorithm, and resampled to a voxel size of 3 × 3 × 3 mm3. Finally, the normalized images were smoothed using a Gaussian kernel with full‐width at half‐maximum (FWHM) of 6 mm.

2.4.2. Static function features

For static brain function, two aspects of measurements were used: the amplitude of low frequency fluctuations (ALFF) for evaluation of regional activity, and the functional connectivity between independent components (IC‐FC) for evaluation of brain connectivity.

Regional ALFF were based on voxel‐level ALFF maps and brain region atlas. Firstly, several nuisance variables including Friston's 24 parameters of head motion (Friston et al., 1996), and signals from global signal, white matter and CSF were regressed out from the preprocessed time series. Then the time series were converted to frequency domain using fast Fourier transform, and the power spectrum was obtained subsequently. The square root of the power spectrum was then calculated and averaged across 0.01–0.1 Hz for each voxel, which was considered as ALFF (Zang et al., 2007). The resulting ALFF map of each subject was converted to z‐score by subtracting the global mean and dividing the global standard deviation for standardization purpose. To define the ROI, the whole brain was segmented into 112 (56 per hemisphere) cortical and subcortical regions by employing a Harvard‐Oxford Structural Atlas (HOA) template (Smith et al., 2004). For each subject, the mean z‐transformed ALFF value of each ROI was extracted as regional ALFF.

Static FC was evaluated using correlation between functionally independent components (IC). The preprocessed fMRI data were decomposed into functional networks by applying a spatial group ICA using Group ICA of fMRI Toolbox (GIFT v4.0b) (https://trendscenter.org/software/gift/). Two data reductions steps were conducted to obtain meaningful ICs. First, 150 principal components (PCs) were obtained by principal component analysis (Erhardt et al., 2011) for each subject data. Second, at group level, the reduced data of all participants were concatenated and further decomposed into 100 PCs using the expectation–maximization algorithm (Himberg et al., 2004). The reliability and stability of the decomposition was ensured by repeating the information‐maximization (Infomax) algorithm for 20 times. Finally, the ICs and corresponding time courses of each subject were derived through back‐reconstruction. Among the 100 ICs, we selected 36 meaningful ICs based on following procedure (Allen et al., 2014): (1) peak activations in gray matter; (2) low spatial overlap with known vascular, ventricular, motion, or susceptibility artifacts; (3) time‐courses dominated by low‐frequency fluctuations and with a high dynamic range. Based on prior knowledge and the spatial correlation values between ICs and given functional network templates, the 36 ICs were classified into the following eight functional subnetworks: anterior and posterior default mode network (aDMN and pDMN), sensorimotor network (SMN), visual network (VN), auditory network (AUN), language network (LN), executive control network (ECN), salience network (SN), and dorsal attention network (DAN). Then additional postprocessing steps were performed on the time courses of the 36 ICs, including: (1) detrending of linear, quadratic, and cubic trends; (2) regressing of six realignment parameters and their temporal derivatives; (3) despiking of detected outliers; and (4) low‐pass filtering with a cut‐off frequency of 0.15 Hz. Finally, the static IC‐FC matrix of each subject was constructed by computing the Pearson correlation coefficient of the postprocessed time courses of 36 ICs, and Fisher's transformation was then performed to improve normality.

2.4.3. Dynamic function features

For dynamic brain function, two measurements including dynamic ALFF (dALFF) and IC dynamic FC (IC‐dFC), in accordance with the two static functional measurements, were used to evaluate the variability or stability of brain activity and connectivity respectively.

We adopted sliding window approach to explore dynamic properties. For each subject, the resting state time courses were divided into windows of 22 TRs in steps of 1 TR, as this decomposition length has been demonstrated to provide a good compromise between the quality of correlation matrix estimation and the ability to resolve dynamics (Allen et al., 2014).

For calculation of regional dALFF, a whole‐brain ALFF map was first constructed for each window, and then the standard deviation (SD) of ALFF across all the windows was calculated for each voxel. Then, the dynamic ALFF (dALFF) value of each of 112 ROIs, same with HOA ROIs in static ALFF analysis, were calculated as the mean ALFF‐SD values of included voxels.

The calculation of IC‐dFC was based on the windows of postprocessed time series for each of the 36 ICs obtained from GICA of static function analysis. For each window, a window‐specific IC‐FC matrix was firstly constructed by computing the Pearson correlation coefficient of the 36 ICs, and Fisher's transformation was then performed to improve normality. Then, IC‐dFC matrix was calculated as the SD values of IC‐FC matrices across all sliding windows, whose element representing the variability of connections between each pair of ICs.

2.5. Selection of MRI features for prediction

For each of the structural and functional features mentioned above, we performed two sample t‐tests to estimate the group differences between MCI_C and MCI_NC groups of SMS cohort when controlling age, gender, and years of education. Multiple comparison corrections were performed: (1) the false discovery rate (FDR) correction was performed for structural parameters (i.e., cortical thickness and subcortical gray matter volume) and the significant level was set to FDR‐corrected p < .05; (2) false positive correction (Lynall et al., 2010), that is, p < (1/N) as significant, was applied for multiple comparison correction of functional measurements involving ALFF/dALFF (N = 112) and IC‐FC/IC‐dFC (N = 36*35/2 = 630). The results with significant group difference were visualized using BrainNet Viewer toolbox (https://www.nitrc.org/projects/bnv/) (Xia et al., 2013).

Finally, the features showing significant difference between MCI_C and MCI_NC patients were selected as the inputs of following prediction models.

2.6. Internal cross‐validation in SMS cohort

In this study, support vector machine (SVM) was utilized to construct single‐modality or multi‐modality prediction models for identification of MCI_C from MCI patients. Briefly, SVM finds an optimal hyperplane which maximizes the margin between the hyperplane and support vectors (Pereira et al., 2009). To evaluate the generalization ability of the proposed feature, a stratified 10‐fold cross‐validation strategy was applied to this study. In this method, the subjects were divided into a training set (90% subjects) and a test set (10% subjects) without changing the proportion of MCI_C in the origin dataset, and repeating this 10 times such that all subjects were part of the test set once. Notably, standard scaling was applied to all the features by subtracting the global mean and dividing the global standard deviation for normalization in SVM algorithm. Then the prediction performance is computed by averaging the quantitative metrics of 10 test sets: accuracy (ACC), sensitivity (SEN), specificity (SPE), and the area under the receiver operating characteristic (ROC) curve (AUC).

Besides, the prediction models using only clinical information (age, gender, years of education, MMSE score, APOE E4 status) and their combination with MRI features, were also constructed using SVM and evaluated. In addition, for construction of all SVM models, both the linear and non‐linear kernel were tested, and the one with better balanced accuracy (bACC), that is, the arithmetic mean of sensitivity and specificity, was selected. An AUC above 0.75 or ACC/SEN/SPE above 70% corresponds to good performance, and AUC above 0.85 or ACC/SEN/SPE above 80% is an excellent performance. In addition, random forest (RF) and logistic regression (LR) were also utilized in this study for comparison, which were evaluated by the same quantitative metrics as SVM.

2.7. External validation in ADNI cohort

The calculation of structural, static, and dynamical regional functional features of ADNI cohort was mainly consistent with that of SMS cohort. Particularly, to obtain the ICs of ADNI cohort corresponding to SMS cohort, instead of GICA, we used the spatially constrained independent component analysis (ICA) algorithm, called group‐information guided independent component analysis (GIG‐ICA), to back reconstruct individual spatial information and time‐courses of 36 ICA components (Du & Fan, 2013) of ADNI participants. Specifically, the 36 ICs derived from SMS cohort were utilized as spatial priori to decompose individual data of ADNI with a multi‐objective optimization strategy which could yield the results with greater component independence and correspondence, and higher spatial and temporal accuracy. GIG‐ICA can generate subject‐specific ICs with stronger independence and better spatial correspondence across subjects, thus avoid potential spatial biases induced by inter‐dataset differences.

The same MCI features for MCI prediction selected by SMS cohort were applied to ADNI cohort, to evaluate the effectiveness of these selected features on an independent external dataset. SVM‐based prediction model construction and evaluation, and the 10‐fold cross‐validation approach was also utilized in ADNI data, in consistency with SMS cohort.

3. RESULTS

3.1. Clinical data

No significant group difference was found between MCI_C and MCI_NC with respect to age (p = .488/.997), gender (p = .287/.206) or years of education (p = .174/.760) at baseline, for both SMS and ADNI cohorts. However, MCI_C group has significantly lower MMSE score and more APOE E4 carriers than MCI_NC group in SMS cohort, but not in ADNI cohort (Table 1).

TABLE 1.

Clinical information of MCI_C and MCI_NC patients in SMS and ADNI cohorts.

Cohort MCI_C MCI_NC T/Chi‐square p
SMS cohort N 41 75 ‐ ‐
Gender (M:F) 16:25 37:38 1.135 .287
Age (years) 72.07 ± 6.67 69.27 ± 7.77 0.836 .488
Education 13.41 ± 2.78 12.16 ± 3.24 1.105 .174
MMSE score 25.37 ± 1.76 26.99 ± 1.73 −4.433 <.001
APOE E4 carrier/non carrier 24:17 18:57 13.689 <.001
ADNI cohort N 25 77 ‐ ‐
Gender (M:F) 14:11 38:39 −1.272 .206
Age (years) 72.48 ± 6.26 72.71 ± 7.12 0.003 .997
Education 15.60 ± 2.53 16.26 ± 2.12 −0.305 .760
MMSE score 27.60 ± 1.68 27.95 ± 1.47 −1.253 .213
APOE E4 carrier/non carrier 12:13 35:42 0.446 .657
Open in a new tab

Abbreviations: ADNI, Alzheimer's Disease Neuroimaging Initiative; APOE, apolipoprotein E; MCI, mild cognitive impairment; MMSE, mini‐mental state examination; SMS, Shanghai Memory Study.

3.2. Structural features

The sMRI‐based structural analysis showed significant reduction of cortical thickness involving temporal cortex (bilateral entorhinal, right middle, and inferior temporal), parietal lobe (bilateral inferior parietal and precuneus, left supramarginal, and right superior parietal) and right caudal middle frontal gyrus, and reduction of subcortical gray matter volume in medial temporal nuclei (bilateral hippocampus and amygdala) and left accumenbens, in MCI_C compared to MCI_NC at baseline (Figure 2).

FIGURE 2.

FIGURE 2

Open in a new tab

Structural differences between MCI_C and MCI_NC. MCI_C showed significantly decreased cortical thickness and subcortical gray matter volume compared to MCI_NC. Results are shown by the segmented ROIs overlaid on the brain surface and coronal images of a representative subject.

3.3. Static function features

Figure 3 showed the significant group differences of static brain function features, that is, ALFF for local activity and IC‐FC for connectivity, between MCI_C and MCI_NC groups.

FIGURE 3.

FIGURE 3

Open in a new tab

Static function differences between MCI_C and MCI_NC. (a) The average ALFF maps of each group and group comparison results of regional ALFF. MCI_C showed significantly increased ALFF of right LOCinf and right AMYG and decreased ALFF of left Pall compared to MCI_NC. **p < .005. (b) The average IC‐FC matrix and group comparison results. MCI_C showed decreased FC between several ICs pairs from different functional networks, which were shown with different colors. AMYG, amygdala; dACC, dorsal anterior cingulate cortex; IPS, intraparietal sulcus; LOCinf, inferior lateral occipital cortex; Pall, pallidum; PCUN, precuneus; PoCG, postcentral gyrus; PP, planum polare; SMG, supramarginal gyrus.

With respect to regional functional activity, MCI_C group showed significantly increased ALFF in right inferior lateral occipital cortex and amygdala, and decreased ALFF in left pallidum, compared to MCI_NC group (Figure 3a).

Group comparison of static brain connectivity revealed significantly decreased inter‐network FC between ICs of supramarginal gyrus (SMG) and intraparietal sulcus (IPS) and dorsal anterior cingulate cortex, and between Planum Polare and postcentral and precuneus in MCI_C group compared to MC_NC (Figure 3b).

3.4. Dynamic function features

Figure 4 showed the significant group differences of dynamic brain function features, that is, dALFF for local activity dynamics and IC‐dFC for brain connectional dynamics, between MCI_C and MCI_NC groups.

FIGURE 4.

FIGURE 4

Open in a new tab

Dynamic function differences between MCI_C and MCI_NC. (a) The average dALFF maps of each group and group comparison results of regional dALFF. MCI_C showed significantly increased dALFF of right inferior lateral occipital cortex, right amygdala, right hippocampus and left anterior parahippocampal gyrus compared to MCI_NC. *p < .05; **p < .005. (b) The average IC‐dFC matrix and group comparison results. Increased inter‐network dFC (red lines) and decreased intra‐network dFC (blue line) was found in MCI_C when compared to MCI_NC. Different functional networks were represented with different colored clusters and characters. AMYG, amygdala; HIP, hippocampus; LOCinf, lateral occipital cortex, inferior division; OP, occipital pole; PCC, post cingulate cortex; PCUN, precuneus; PHGant, anterior parahippocampal gyrus; SMG, supramarginal gyrus.

The whole‐brain dALFF maps of MCI_C and MCI_NC groups showed similar distributions. Group comparison revealed that the MCI_C patients had significantly increased dALFF in left anterior parahippocampal gyrus, right amygdala, hippocampus, and LOCinf, as compared to MCI_NC group (Figure 4a).

With respect to dynamical functional connectivity, as illustrated in Figure 4b, MCI_C group demonstrated increased inter‐network dFC between SMG and occipital pole, and decreased intra‐network dFC between post cingulate cortex (PCC) and precuneus.

3.5. Internal cross‐validation in SMS cohort

Based on the selected MRI features showing differences between MCI converter and non‐converters shown above (i.e., 16 sMRI, 7 static fMRI, and 7 dynamic fMRI features) (Figures 2, 3, 4), SVM‐based prediction models of MCI conversion were constructed and evaluated using 10‐fold cross‐validation in SMS cohort, as shown in Figure 5a and Table 2.

FIGURE 5.

FIGURE 5

Open in a new tab

The ROC curves of the prediction models in internal SMS cohort (a) and external ADNI cohort (b).

TABLE 2.

Accuracy (ACC), sensitivity (SEN), specificity (SPE), and area under the curve (AUC) of support vector machine‐based prediction models for mild cognitive impairment conversion in internal Shanghai Memory Study (SMS) cohort and external Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.

SMS cohort (internal cross‐validation) ADNI cohort (external validation)
AUC ACC (%) SEN (%) SPE (%) AUC ACC (%) SEN (%) SPE (%)
sMRI 0.81 75.9 73.0 77.1 0.82 77 76.7 76.9
fMRI‐static 0.80 72.3 62.5 77.5 0.74 72.1 61.7 75.2
fMRI‐dynamic 0.75 71.3 62.5 75.9 0.77 74.2 56.7 79.8
fMRI graphic file with name HBM-45-e26529-g019.jpg graphic file with name HBM-45-e26529-g012.jpg graphic file with name HBM-45-e26529-g008.jpg graphic file with name HBM-45-e26529-g005.jpg 0.76 76.4 65.0 79.8
sMRI + fMRI graphic file with name HBM-45-e26529-g022.jpg graphic file with name HBM-45-e26529-g004.jpg graphic file with name HBM-45-e26529-g002.jpg graphic file with name HBM-45-e26529-g020.jpg graphic file with name HBM-45-e26529-g014.jpg graphic file with name HBM-45-e26529-g021.jpg graphic file with name HBM-45-e26529-g006.jpg graphic file with name HBM-45-e26529-g013.jpg
Clinical graphic file with name HBM-45-e26529-g003.jpg 77.4 graphic file with name HBM-45-e26529-g018.jpg 76.1 0.62 71.8 60.0 76.3
MRI + clinical graphic file with name HBM-45-e26529-g015.jpg graphic file with name HBM-45-e26529-g009.jpg graphic file with name HBM-45-e26529-g001.jpg graphic file with name HBM-45-e26529-g007.jpg graphic file with name HBM-45-e26529-g023.jpg 79.0 78.3 78.9
Open in a new tab

Note: AUC above 0.75 or ACC/SEN/SPE above 70% are shown in bold to indicate good performances, and AUC above 0.85 or ACC/SEN/SPE above 80% are indicated by rectangles to indicate excellent performances.

Abbreviations: fMRI, functional magnetic resonance imaging; sMRI, structural magnetic resonance imaging.

Among all the imaging‐based prediction models, the fMRI involving both static and dynamic features (fMRI model) and their combination with sMRI (sMRI + fMRI, i.e., all selected MRI features in this study) revealed best performances with ACC/sensitivity/specificity of 81%–95% and AUC above 0.9, followed by the sMRI model showing good performance with ACC/sensitivity/specificity above 73% and AUC of 0.81. Relatively, the static fMRI or dynamic fMRI solely resulted in an unsatisfactory sensitivity that below 70%, while the AUC, AUC, and specificity were similarly good with sMRI. Besides, when the MRI features were combined with clinical parameters, the mixed imaging‐clinical model (MRI + clinical) performed slightly better than the integrated imaging model (MRI, i.e., sMRI + fMRI) as indicated by all indices.

3.6. External validation in ADNI cohort

The external validation of the SMS‐derived features was performed in the ADNI cohort, as shown in Figure 5b and Table 2.

Similar with SMS cohort, the sMRI, fMRI and multi‐modality MRI (i.e., sMRI + fMRI) showed overall good or excellent performance with ACC/sensitivity/specificity of 70%–80% and AUC above 0.74 in ADNI cohort. Among them, the MRI model reached an excellent performance with highest AUC = 0.86, ACC = 81.6%, sensitivity = 83.3%, and specificity = 80.7%. And in consistency with SMS cohort, while the static fMRI and dynamic fMRI model solely showed relatively unsatisfactory sensitivity below 70%, their combination (i.e., the fMRI model) performed generally good with an increase in sensitivity and ACC.

In addition, RF‐based and LR‐based prediction models for MCI conversion further demonstrated the superiority of multi‐modality MRI models among all imaging models, but with an overall worse performance than SVM classifier (Figure S2, Tables S1 and S2).

4. DISCUSSION

In this study, we proposed a novel framework by integrating structural MRI and static and dynamic fMRI, detected a series of effective features for machine‐learning‐based prediction of MCI conversion to AD, and validated them in an extremely different external cohort. The key findings of this study are as follows: (1) The MRI predictors for MCI conversion were mainly characterized by alterations of medial temporal lobe (MTL) with atrophy extending to lateral temporal and regional hyperactivity and instability, posterior parietal cortex (PPC) with atrophy and inter‐regional hypo‐connectivity and connectional instability, and occipital cortex with functional instability; (2) The proposed analysis framework revealed effective MRI‐based features for effective prediction of MCI conversion, with AUC above 0.7 and ACC above 70% for all prediction models in both SMS and external ADNI cohort, despite of the prominent inter‐cohort differences; (3) The combination of static and dynamic fMRI features resulted in overall good performance as relative to the un‐satisfactory sensitivity of static or dynamic fMRI solely, and the incorporation of all MRI features provided excellent performance with AUC above 0.85 and average ACC/sensitivity/specificity around 80%, in both cohorts.

4.1. Brain MRI predictors for MCI conversion

The proposed MRI predictors indicate a specific structural and functional pattern before MCI conversion to AD, some of which in consistent with the reported neurodegeneration process of AD, while others suggesting novel features and earlier dysfunction in MCI converters before AD diagnosis, providing a new insight into the mechanisms approaching the time point of conversion.

The MTL, including the hippocampus, amygdala, entorhinal cortex, and parahippocampal gyrus regions involved in this study, has been widely implicated in prodromal and early AD. In consistency with the Braak's neuropathological staging of AD, the loss of volume in entorhinal cortex and hippocampus was reported to happen earliest, and has been used as biomarkers of AD conversion in MCI patients (Braak & Braak, 1991; Chupin et al., 2009; Clerx et al., 2013; Leandrou et al., 2018). In this study, we found that the structural biomarkers of MCI conversion were not restricted in MTL, but with an extension to lateral temporal regions. Specifically, our results demonstrated alterations of both two functional aspects (the regional hyperactivity and the increased instability) in addition to decreased volumes, in amygdala rather than other regions in MCI converters, suggesting an important role of amygdala for prediction of MCI conversion.

The PPC is a polymodal area, including precuneus and PCC in medial lobe, and SMG, IPS, superior and inferior parietal lobule (SPL/IPL) involved in our results (Jacobs et al., 2012). In our results, notably, the alterations of PPC area in MCI_C showed a unique and characterized pattern mainly involving atrophy and inter‐regional connectional profiles, but not regional/local activities. Different PPC regions contribute to separate but associated functions through complex inter‐connected brain networks (Sestieri et al., 2017). The precuneus and PCC, the hubs of posterior default mode network (pDMN), may play an important role in tau pathology of parietal lobe, which is associated with the spread of tau from entorhinal cortex via the connectivity between hippocampus and pDMN regions (Khan et al., 2014; Ziontz et al., 2021). Dysfunction of pDMN has been consistently reported in studies of AD (Badhwar et al., 2017), and was proved to relate with episodic memory (Khan et al., 2014; Ziontz et al., 2021) which is the major impaired function in early AD (Hodges, 2006), and visual memory (Sperling et al., 2003) that is also impaired in AD and MCI patients (Didic et al., 2013; Quental et al., 2013). Accordingly, the atrophy in precuneus may correspond to deteriorated memory function in MCI converters, while its hypo‐connectivity with temporal region (by IC‐FC) and reduced connectional complexity with another DMN node PCC (by IC‐dFC) suggest a segregated state or declined integration (Ishaque et al., 2017) of DMN with other networks before conversion to AD. The SPL and IPS are the key regions of dorsal attention network (DAN) important for the up‐down modulation to sensory regions (Sestieri et al., 2017), which may be disturbed in MCI converters as shown by SPL atrophy and hypo‐connection between IPS and the sensorimotor network node SMG in our results. And alterations of SMG in MCI converters, including atrophy and inter‐region functional alterations, are more likely to reflect brain changes corresponding to phonological decision processing and verbal working memory (Deschamps et al., 2014; Hartwigsen et al., 2010), functions that are usually not impaired early in AD continuum but rather follow the wake of episodic memory and attention symptoms (Hodges, 2006).

Occipital cortex is pathologically recognized to be involved at late stage of AD (Braak & Braak, 1991). Several recent fMRI studies have reported the involvement of occipital cortex in AD including decreased functional centrality and DMN‐occipital connection (Binnewijzend et al., 2014; Li et al., 2016) and also decreased dALFF in predementia AD (Li et al., 2020). In this study, we did not find atrophy of occipital areas in MCI converters, but detected functional alterations, particularly an increased regional and connectional instability. The connectional instability between occipital pole and SMG in MCI_C patients suggested a change of the brain dynamics along the dorsal visual pathway (parieto‐occipital stream), which is involved in visuospatial function and is commonly called “where” system for processing spatial location (Yamasaki et al., 2012). Besides, the lateral occipital cortex (LOC), showing regional hyperactivity and instability in MCI converters, is a critical region for shape perception yielding greater activation when viewing intact objects than scrambled ones (Malach et al., 1995). Overall, these results indicate an earlier occurrence of occipital dysfunction characterized predominantly by dynamic features before MCI conversion.

4.2. Inter‐cohort effective MRI features

Validation of MRI features in an external cohort is rare in AD‐related MRI studies, few for predicting MCI conversion (Bron et al., 2021; Hall et al., 2015). Bron et al. constructed the MCI prediction model based on voxel‐level features of T1 MRI from 859 cases of ADNI, and tested in 139 cases of Netherlands cohort with an AUC of 0.702 and ACC of 64.7% (Bron et al., 2021). Hall et al. tested the sMRI features and a disease state index integrating sMRI and clinical data across four datasets (from 123 to 370 subjects in each dataset), revealing AUC of 0.71–0.75, ACC of 65%–75%, sensitivity of 54%–74%, specificity of 58%–77% for sMRI features only, and AUC of 0.72–0.76, ACC of 66%–79%, sensitivity of 56%–70%, specificity of 67%–81% for the integrated index (Hall et al., 2015). In this study, although using a smaller number of patients' data, our prediction model based on ROI‐wise sMRI features of SMS cohort showed a similar performance with higher AUC in external cohort (AUC = 0.82, ACC = 77%, sensitivity = 76.7%, specificity = 76.9%), demonstrating the robustness of sMRI in MCI conversion prediction and the efficiency of the selected ROI features.

Establishment of effective fMRI predictors across different cohorts is quite challenging and lacks in AD‐related studies. By applying different fMRI‐based analysis methods to predict MCI conversion, several studies explored functional differences between MCI converters and non‐converters (Delli Pizzi et al., 2019; Deng et al., 2016; Qiu et al., 2016), while others also tried to establish a prediction model or tested the performance (Gupta et al., 2020; Hojjati et al., 2017, 2018; Li et al., 2016), based on single dataset with limited patients and showing substantial variation between results. Recently, de Vos et al. evaluated the performance of a variety of fMRI measures for distinguishing of AD from HC, and found that the ICA‐based FC matrices, FC dynamic and ALFF are most discriminative, with larger AUC than graph properties, FC states and voxel‐based FC or centrality (de Vos et al., 2018). Inspired by this finding, for predicting of MCI conversion, we also extracted MRI features using an ICA‐based strategy. Besides, we additionally include a dALFF measure to evaluate the regional functional dynamics, and using a ROI‐wise rather than voxel‐wise calculation for inter‐cohort consistency and convenience of future clinic use. Based on these fMRI features, the fMRI model showed a good performance for MCI conversion prediction similar with sMRI on external ADNI dataset.

Our analysis strategy takes the population heterogeneity of AD into consideration. Specifically, the models constructed from SMS cohort were tuned when validated on ADNI cohort, using same selected MRI predictors as inputs. Some studies have proved variations of neurodegeneration process across different ethnoracial groups in AD, although with common pathological basis. For example, a post‐mortem study found the Black decedents had lower Braak tangle stage compared to Hispanic decedents and white decedents (Santos et al., 2019). Recently, a study on neuroimaging biomarkers of AD for racial differences has unveiled that Africa American had decreased AD signature volume than Whites (Meeker et al., 2021). Moreover, the genome‐wide alterations between Chinese and Caucasians of AD patients have been proposed in 2020 (Qin et al., 2020). Given these factors in nature, it is plausible to tune the neuroimaging models to obtain population‐specific parameters for the selected MRI predictors, which alleviate the influence of population heterogeneity and contribute to the effectiveness of proposed imaging features.

4.3. Performance of prediction models

The fMRI model combining static and dynamic functional features achieved good‐to‐excellent performance in SMS and ADNI cohorts, demonstrating the advantage of investigating both static and dynamic profiles for MCI conversion prediction. The multi‐modality model, with AUC above 0.85 and average ACC/sensitivity/specificity around 80% in both cohorts, showed superiority to sMRI or fMRI solely, which is consistent with conclusion of several previous studies also suggesting better performance of multi‐modality than single‐modality model for MCI conversion prediction (Gupta et al., 2020; Hojjati et al., 2018) or AD classification (Schouten et al., 2016). While the proposed analysis framework and MRI predictors integrate structural, static, and dynamic functional features targeting different aspects of brain, the effectiveness of the models also benefits from SVM, a powerful machine learning methodology who maps the samples to high‐dimensional feature space using linear or non‐linear kernels and identifies the optimal hyper plane for classification.

The results of clinical‐feature‐based models were substantially different between SMS and ADNI cohorts, with satisfactory performance on SMS and lowest AUC = 0.62 in ADNI. The inter‐dataset clinical differences may have an important impact. In SMS cohort, MCI converters presented severer cognitive impairment with lower MMSE score, a higher percentage of patients carrying APOE E4 than MCI non‐converters. In contrast, there was no clinical difference between MCI converters and non‐converters in ADNI cohort. Consequently, the imaging‐clinical model obtained an improved prediction capability than imaging model on SMS, supporting the contribution of clinical features, but unimproved results on ADNI with even lower sensitivity. Although whether the inter‐dataset clinical differences are occasional or due to population variations could not be decided from this study, these results demonstrate generally better performances of imaging‐based models than clinical features.

4.4. Limitations and future work

Several limitations should be stated. First, due to the lack of longitudinal datasets with both fMRI data and conversion outcome, only two cohorts, that is, SMS with 116 patients (41 converters) and public cohort ADNI with 102 patients (25 converters) were used for internal and external validation of the proposed framework and features. If more longitudinal data (especially converters) should be available in the future, more ideal inter‐cohort validation method such as swapping the two datasets would be performed to test the universality of the framework and MRI features. Further tests are needed before clinical use could eventually become a reality. Second, current study was based on the MRI technique. PET imaging or CSF biomarkers has been proved to have excellent prediction performance in MCI conversion prediction (Huang et al., 2020; Zhao et al., 2020), which may improve the performance of MRI‐based models, which could be tested in future.

5. CONCLUSIONS

In conclusion, this inter‐cohort validation study proposed a novel analysis framework and detected effective features with good application to extremely different external cohort, providing a new insight into the mechanisms of MCI converters and paving a way for clinical use of MRI biomarkers in future.

AUTHOR CONTRIBUTIONS

Liqin Yang and Qianhua Zhao contributed to the conception of the study, designed the experiment, and revised the manuscript. Zhihan Chen performed the data analysis and drafted and revised the manuscript. Keliang Chen and Yuxin Li contributed to the clinical diagnosis, data acquisition, and provided critical comments and suggestions. Daoying Geng, Xiantao Li, Huimeng Lu, Xiaoniu Liang, Saineng Ding, Zhenxu Xiao, Xiaoxi Ma, Li Zheng, and Ding Ding contributed to clinical data acquisition. All authors read and approved the final manuscript.

FUNDING INFORMATION

This work was supported by the National Natural Science Foundation of China (82102132, 82071200, 82001139), the Science and Technology Commission of Shanghai Municipality (20S31904300), the National Project of Chronic Disease of China (2016YFC1306402), the Clinical Research Plan of Shanghai Hospital Development Center (SHDC2020CR4007), Shanghai Municipal Key Clinical Specialty (shslczdzk03201).

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflict of interest.

Supporting information

Data S1. Supporting Information.

Click here for additional data file. (324KB, docx)

ACKNOWLEDGMENTS

Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH‐12‐2‐0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol‐Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann‐La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California. We also thankfully acknowledge participation of patients.

Chen, Z. , Chen, K. , Li, Y. , Geng, D. , Li, X. , Liang, X. , Lu, H. , Ding, S. , Xiao, Z. , Ma, X. , Zheng, L. , Ding, D. , Zhao, Q. , Yang, L. , & for the Alzheimer's Disease Neuroimaging Initiative (2024). Structural, static, and dynamic functional MRI predictors for conversion from mild cognitive impairment to Alzheimer's disease: Inter‐cohort validation of Shanghai Memory Study and ADNI . Human Brain Mapping, 45(1), e26529. 10.1002/hbm.26529

Zhihan Chen, Keliang Chen, and Yuxin Li contributed equally to this work.

Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.

Contributor Information

Qianhua Zhao, Email: qianhuazhao@fudan.edu.cn.

Liqin Yang, Email: mozi001_ren@sina.com.

DATA AVAILABILITY STATEMENT

The part of datasets supporting the conclusions of this article is available in Alzheimer's Disease Neuroimaging Initiative (ADNI) data repository, http://adni.loni.usc.edu. The Freesurfer version 6.0.0 can be freely downloaded from http://surfer.nmr.mgh.harvard.edu/. DPABI is freely available at http://rfmri.org/dpabi. Group ICA of fMRI Toolbox (GIFT v3.0b) is freely available at http://mialab.mrn.org/software/gift. The BrainNet Viewer is freely available on the NITRC website: http://www.nitrc.org/projects/bnv.

REFERENCES

  1. Albert, M. S. , DeKosky, S. T. , Dickson, D. , Dubois, B. , Feldman, H. H. , Fox, N. C. , Gamst, A. , Holtzman, D. M. , Jagust, W. J. , Petersen, R. C. , Snyder, P. J. , Carrillo, M. C. , Thies, B. , & Phelps, C. H. (2011). The diagnosis of mild cognitive impairment due to Alzheimer's disease: Recommendations from the National Institute on Aging‐Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 7, 270–279. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Allen, E. A. , Damaraju, E. , Plis, S. M. , Erhardt, E. B. , Eichele, T. , & Calhoun, V. D. (2014). Tracking whole‐brain connectivity dynamics in the resting state. Cerebral Cortex, 24, 663–676. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Ansart, M. , Epelbaum, S. , Bassignana, G. , Bone, A. , Bottani, S. , Cattai, T. , Couronne, R. , Faouzi, J. , Koval, I. , Louis, M. , Thibeau‐Sutre, E. , Wen, J. , Wild, A. , Burgos, N. , Dormont, D. , Colliot, O. , & Durrleman, S. (2021). Predicting the progression of mild cognitive impairment using machine learning: A systematic, quantitative and critical review. Medical Image Analysis, 67, 101848. [DOI] [PubMed] [Google Scholar]
  4. Arbabshirani, M. R. , Plis, S. , Sui, J. , & Calhoun, V. D. (2017). Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls. NeuroImage, 145, 137–165. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Badhwar, A. , Tam, A. , Dansereau, C. , Orban, P. , Hoffstaedter, F. , & Bellec, P. (2017). Resting‐state network dysfunction in Alzheimer's disease: A systematic review and meta‐analysis. Alzheimer's & Dementia (Amsterdam), 8, 73–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Binnewijzend, M. A. , Adriaanse, S. M. , Van der Flier, W. M. , Teunissen, C. E. , de Munck, J. C. , Stam, C. J. , Scheltens, P. , van Berckel, B. N. , Barkhof, F. , & Wink, A. M. (2014). Brain network alterations in Alzheimer's disease measured by eigenvector centrality in fMRI are related to cognition and CSF biomarkers. Human Brain Mapping, 35, 2383–2393. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Biswal, B. , Yetkin, F. Z. , Haughton, V. M. , & Hyde, J. S. (1995). Functional connectivity in the motor cortex of resting human brain using echo‐planar MRI. Magnetic Resonance in Medicine, 34, 537–541. [DOI] [PubMed] [Google Scholar]
  8. Braak, H. , & Braak, E. (1991). Neuropathological staging of Alzheimer‐related changes. Acta Neuropathologica, 82, 239–259. [DOI] [PubMed] [Google Scholar]
  9. Bron, E. E. , Klein, S. , Papma, J. M. , Jiskoot, L. C. , Venkatraghavan, V. , Linders, J. , Aalten, P. , De Deyn, P. P. , Biessels, G. J. , Claassen, J. , Middelkoop, H. A. M. , Smits, M. , Niessen, W. J. , van Swieten, J. C. , van der Flier, W. M. , Ramakers, I. , & van der Lugt, A. (2021). Cross‐cohort generalizability of deep and conventional machine learning for MRI‐based diagnosis and prediction of Alzheimer's disease. Neuroimage: Clinical, 31, 102712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Busse, A. , Hensel, A. , Gühne, U. , Angermeyer, M. C. , & Riedel‐Heller, S. G. (2006). Mild cognitive impairment long‐term course of four clinical subtypes. Neurology, 67, 2176–2185. [DOI] [PubMed] [Google Scholar]
  11. Calhoun, V. D. , & de Lacy, N. (2017). Ten key observations on the analysis of resting‐state functional MR imaging data using independent component analysis. Neuroimaging Clinics of North America, 27, 561–579. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Chupin, M. , Gérardin, E. , Cuingnet, R. , Boutet, C. , Lemieux, L. , Lehéricy, S. , Benali, H. , Garnero, L. , Colliot, O. , & Alzheimer's Disease Neuroimaging Initiative . (2009). Fully automatic hippocampus segmentation and classification in Alzheimer's disease and mild cognitive impairment applied on data from ADNI. Hippocampus, 19, 579–587. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Clerx, L. , van Rossum, I. A. , Burns, L. , Knol, D. L. , Scheltens, P. , Verhey, F. , Aalten, P. , Lapuerta, P. , van de Pol, L. , van Schijndel, R. , de Jong, R. , Barkhof, F. , Wolz, R. , Rueckert, D. , Bocchetta, M. , Tsolaki, M. , Nobili, F. , Wahlund, L.‐O. , Minthon, L. , … Visser, P. J. (2013). Measurements of medial temporal lobe atrophy for prediction of Alzheimer's disease in subjects with mild cognitive impairment. Neurobiology of Aging, 34, 2003–2013. [DOI] [PubMed] [Google Scholar]
  14. Cordova‐Palomera, A. , Kaufmann, T. , Persson, K. , Alnaes, D. , Doan, N. T. , Moberget, T. , Lund, M. J. , Barca, M. L. , Engvig, A. , Braekhus, A. , Engedal, K. , Andreassen, O. A. , Selbaek, G. , & Westlye, L. T. (2017). Disrupted global metastability and static and dynamic brain connectivity across individuals in the Alzheimer's disease continuum. Scientific Reports, 7, 40268. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. de Vos, F. , Koini, M. , Schouten, T. M. , Seiler, S. , van der Grond, J. , Lechner, A. , Schmidt, R. , de Rooij, M. , & Rombouts, S. A. R. B. (2018). A comprehensive analysis of resting state fMRI measures to classify individual patients with Alzheimer's disease. NeuroImage, 167, 62–72. [DOI] [PubMed] [Google Scholar]
  16. Delli Pizzi, S. , Punzi, M. , Sensi, S. L. , & Alzheimer's Disease Neuroimaging Initiative . (2019). Functional signature of conversion of patients with mild cognitive impairment. Neurobiology of Aging, 74, 21–37. [DOI] [PubMed] [Google Scholar]
  17. Deng, Y. , Liu, K. , Shi, L. , Lei, Y. , Liang, P. , Li, K. , Chu, W. C. W. , Wang, D. , & Alzheimer's Disease Neuroimaging Initiative . (2016). Identifying the alteration patterns of brain functional connectivity in progressive mild cognitive impairment patients: A longitudinal whole‐brain voxel‐wise degree analysis. Frontiers in Aging Neuroscience, 8, 195. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Deschamps, I. , Baum, S. R. , & Gracco, V. L. (2014). On the role of the supramarginal gyrus in phonological processing and verbal working memory: Evidence from rTMS studies. Neuropsychologia, 53, 39–46. [DOI] [PubMed] [Google Scholar]
  19. Desikan, R. S. , Segonne, F. , Fischl, B. , Quinn, B. T. , Dickerson, B. C. , Blacker, D. , Buckner, R. L. , Dale, A. M. , Maguire, R. P. , Hyman, B. T. , Albert, M. S. , & Killiany, R. J. (2006). An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage, 31, 968–980. [DOI] [PubMed] [Google Scholar]
  20. Didic, M. , Felician, O. , Barbeau, E. J. , Mancini, J. , Latger‐Florence, C. , Tramoni, E. , & Ceccaldi, M. (2013). Impaired visual recognition memory predicts Alzheimer's disease in amnestic mild cognitive impairment. Dementia and Geriatric Cognitive Disorders, 35, 291–299. [DOI] [PubMed] [Google Scholar]
  21. Du, Y. , & Fan, Y. (2013). Group information guided ICA for fMRI data analysis. NeuroImage, 69, 157–197. [DOI] [PubMed] [Google Scholar]
  22. Erhardt, E. B. , Rachakonda, S. , Bedrick, E. J. , Allen, E. A. , Adali, T. , & Calhoun, V. D. (2011). Comparison of multi‐subject ICA methods for analysis of fMRI data. Human Brain Mapping, 32, 2075–2095. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Filippi, M. , Spinelli, E. G. , Cividini, C. , & Agosta, F. (2019). Resting state dynamic functional connectivity in neurodegenerative conditions: A review of magnetic resonance imaging findings. Frontiers in Neuroscience, 13, 657. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Fischl, B. (2012). FreeSurfer. Neuroimage, 62, 774–781. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Friston, K. J. , Williams, S. , Howard, R. , Frackowiak, R. S. J. , & Turner, R. (1996). Movement‐related effects in fMRI time‐series. Magnetic Resonance in Medicine, 35, 346–355. [DOI] [PubMed] [Google Scholar]
  26. Gupta, Y. , Kim, J. I. , Kim, B. C. , & Kwon, G. R. (2020). Classification and graphical analysis of Alzheimer's disease and its prodromal stage using multimodal features from structural, diffusion, and functional neuroimaging data and the APOE genotype. Frontiers in Aging Neuroscience, 12, 238. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hall, A. , Munoz‐Ruiz, M. , Mattila, J. , Koikkalainen, J. , Tsolaki, M. , Mecocci, P. , Kloszewska, I. , Vellas, B. , Lovestone, S. , Visser, P. J. , Lotjonen, J. , Soininen, H. , Alzheimer Disease Neuroimaging Initiative , AddNeuroMed consortium , & DESCRIPA and Kuopio L‐MCI . (2015). Generalizability of the disease state index prediction model for identifying patients progressing from mild cognitive impairment to Alzheimer's disease. Journal of Alzheimer's Disease, 44, 79–92. [DOI] [PubMed] [Google Scholar]
  28. Hartwigsen, G. , Baumgaertner, A. , Price, C. J. , Koehnke, M. , Ulmer, S. , & Siebner, H. R. (2010). Phonological decisions require both the left and right supramarginal gyri. Proceedings of the National Academy of Sciences of the United States of America, 107, 16494–16499. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Himberg, J. , Hyvärinen, A. , & Esposito, F. (2004). Validating the independent components of neuroimaging time series via clustering and visualization. NeuroImage, 22, 1214–1222. [DOI] [PubMed] [Google Scholar]
  30. Hodges, J. R. (2006). Alzheimer's centennial legacy: Origins, landmarks and the current status of knowledge concerning cognitive aspects. Brain, 129, 2811–2822. [DOI] [PubMed] [Google Scholar]
  31. Hohenfeld, C. , Werner, C. J. , & Reetz, K. (2018). Resting‐state connectivity in neurodegenerative disorders: Is there potential for an imaging biomarker? Neuroimage: Clinical, 18, 849–870. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Hojjati, S. H. , Ebrahimzadeh, A. , Khazaee, A. , Babajani‐Feremi, A. , & Alzheimer's Disease Neuroimaging Initiative . (2017). Predicting conversion from MCI to AD using resting‐state fMRI, graph theoretical approach and SVM. Journal of Neuroscience Methods, 282, 69–80. [DOI] [PubMed] [Google Scholar]
  33. Hojjati, S. H. , Ebrahimzadeh, A. , Khazaee, A. , Babajani‐Feremi, A. , & Alzheimer's Disease Neuroimaging Initiative . (2018). Predicting conversion from MCI to AD by integrating rs‐fMRI and structural MRI. Computers in Biology and Medicine, 102, 30–39. [DOI] [PubMed] [Google Scholar]
  34. Huang, K. , Lin, Y. , Yang, L. , Wang, Y. , Cai, S. , Pang, L. , Wu, X. , Huang, L. , & Alzheimer's Disease Neuroimaging Initiative . (2020). A multipredictor model to predict the conversion of mild cognitive impairment to Alzheimer's disease by using a predictive nomogram. Neuropsychopharmacology, 45, 358–366. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Hutchison, R. M. , Womelsdorf, T. , Allen, E. A. , Bandettini, P. A. , Calhoun, V. D. , Corbetta, M. , Della Penna, S. , Duyn, J. H. , Glover, G. H. , Gonzalez‐Castillo, J. , Handwerker, D. A. , Keilholz, S. , Kiviniemi, V. , Leopold, D. A. , de Pasquale, F. , Sporns, O. , Walter, M. , & Chang, C. (2013). Dynamic functional connectivity: Promise, issues, and interpretations. NeuroImage, 80, 360–378. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Ishaque, M. , Manning, J. H. , Woolsey, M. D. , Franklin, C. G. , Tullis, E. W. , Beckmann, C. F. , & Fox, P. T. (2017). Functional integrity in children with anoxic brain injury from drowning. Human Brain Mapping, 38, 4813–4831. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Jacobs, H. I. , Van Boxtel, M. P. , Jolles, J. , Verhey, F. R. , & Uylings, H. B. (2012). Parietal cortex matters in Alzheimer's disease: An overview of structural, functional and metabolic findings. Neuroscience and Biobehavioral Reviews, 36, 297–309. [DOI] [PubMed] [Google Scholar]
  38. Jones, D. T. , Vemuri, P. , Murphy, M. C. , Gunter, J. L. , Senjem, M. L. , Machulda, M. M. , Przybelski, S. A. , Gregg, B. E. , Kantarci, K. , Knopman, D. S. , Boeve, B. F. , Petersen, R. C. , & Jack, C. R. (2012). Non‐stationarity in the “resting brain's” modular architecture. PLoS One, 7, e39731. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Kam, T. E. , Zhang, H. , Jiao, Z. , & Shen, D. (2020). Deep learning of static and dynamic brain functional networks for early MCI detection. IEEE Transactions on Medical Imaging, 39, 478–487. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Katzman, R. , Zhang, M. Y. , Ouangyaqu , Wang, Z. Y. , Liu, W. T. , Yu, E. , Wong, S. C. , Salmon, D. P. , & Grant, I. (1988). A Chinese version of the mini‐mental state examination – impact of illiteracy in a Shanghai dementia survey. Journal of Clinical Epidemiology, 41, 971–978. [DOI] [PubMed] [Google Scholar]
  41. Khan, U. A. , Liu, L. , Provenzano, F. A. , Berman, D. E. , Profaci, C. P. , Sloan, R. , Mayeux, R. , Duff, K. E. , & Small, S. A. (2014). Molecular drivers and cortical spread of lateral entorhinal cortex dysfunction in preclinical Alzheimer's disease. Nature Neuroscience, 17, 304–311. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Koch, W. , Ehrenhaft, A. , Griesser, K. , Pfeufer, A. , Müller, J. , Schömig, A. , & Kastrati, A. (2002). TaqMan systems for genotyping of disease‐related polymorphisms present in the gene encoding apolipoprotein E. Clinical Chemistry and Laboratory Medicine, 40, 1123–1131. [DOI] [PubMed] [Google Scholar]
  43. Leandrou, S. , Petroudi, S. , Kyriacou, P. A. , Reyes‐Aldasoro, C. C. , & Pattichis, C. S. (2018). Quantitative MRI brain studies in mild cognitive impairment and Alzheimer's disease: A methodological review. IEEE Reviews in Biomedical Engineering, 11, 97–111. [DOI] [PubMed] [Google Scholar]
  44. Lee, M. H. , Smyser, C. D. , & Shimony, J. S. (2013). Resting‐state fMRI: A review of methods and clinical applications. AJNR. American Journal of Neuroradiology, 34, 1866–1872. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Li, Y. , Wang, X. , Li, Y. , Sun, Y. , Sheng, C. , Li, H. , Li, X. , Yu, Y. , Chen, G. , Hu, X. , Jing, B. , Wang, D. , Li, K. , Jessen, F. , Xia, M. , & Han, Y. (2016). Abnormal resting‐state functional connectivity strength in mild cognitive impairment and its conversion to Alzheimer's disease. Neural Plasticity, 2016, 4680972. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Li, Z. , Li, K. , Luo, X. , Zeng, Q. , Zhao, S. , Zhang, B. , Zhang, M. , Chen, Y. , & Alzheimer's Disease Neuroimaging Initiative . (2020). Distinct brain functional impairment patterns between suspected non‐Alzheimer disease pathophysiology and Alzheimer's disease: A study combining static and dynamic functional magnetic resonance imaging. Frontiers in Aging Neuroscience, 12, 550664. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Lynall, M. E. , Bassett, D. S. , Kerwin, R. , McKenna, P. J. , Kitzbichler, M. , Muller, U. , & Bullmore, E. (2010). Functional connectivity and brain networks in schizophrenia. The Journal of Neuroscience, 30, 9477–9487. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Malach, R. , Reppas, J. B. , Benson, R. R. , Kwong, K. K. , Jiang, H. , Kennedy, W. A. , Ledden, P. J. , Brady, T. J. , Rosen, B. R. , & Tootell, R. B. (1995). Object‐related activity revealed by functional magnetic resonance imaging in human occipital cortex. Proceedings of the National Academy of Sciences of the United States of America, 92, 8135–8139. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. McKhann, G. , Drachman, D. , Folstein, M. , Katzman, R. , Price, D. , & Stadlan, E. M. (1984). Clinical diagnosis of Alzheimer's disease – Report of the NINCDS‐ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's disease. Neurology, 34, 939–944. [DOI] [PubMed] [Google Scholar]
  50. Meeker, K. L. , Wisch, J. K. , Hudson, D. , Coble, D. , Xiong, C. , Babulal, G. M. , Gordon, B. A. , Schindler, S. E. , Cruchaga, C. , Flores, S. , Dincer, A. , Benzinger, T. L. , Morris, J. C. , & Ances, B. M. (2021). Socioeconomic status mediates racial differences seen using the AT(N) framework. Annals of Neurology, 89, 254–265. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Moguilner, S. , Garcia, A. M. , Perl, Y. S. , Tagliazucchi, E. , Piguet, O. , Kumfor, F. , Reyes, P. , Matallana, D. , Sedeno, L. , & Ibanez, A. (2021). Dynamic brain fluctuations outperform connectivity measures and mirror pathophysiological profiles across dementia subtypes: A multicenter study. NeuroImage, 225, 117522. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Molinuevo, J. L. , Gómez‐Anson, B. , Monte, G. C. , Bosch, B. , Sánchez‐Valle, R. , & Rami, L. (2011). Neuropsychological profile of prodromal Alzheimer's disease (Prd‐AD) and their radiological correlates. Archives of Gerontology and Geriatrics, 52, 190–196. [DOI] [PubMed] [Google Scholar]
  53. Pereira, F. , Mitchell, T. , & Botvinick, M. (2009). Machine learning classifiers and fMRI: A tutorial overview. NeuroImage, 45, S199–S209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Petersen, R. C. (2004). Mild cognitive impairment as a diagnostic entity. Journal of Internal Medicine, 256, 183–194. [DOI] [PubMed] [Google Scholar]
  55. Petersen, R. C. , Doody, R. , Kurz, A. , Mohs, R. C. , Morris, J. C. , Rabins, P. V. , Ritchie, K. , Rossor, M. , Thal, L. , & Winblad, B. (2001). Current concepts in mild cognitive impairment. Archives of Neurology, 58, 1985–1992. [DOI] [PubMed] [Google Scholar]
  56. Power, J. D. , Mitra, A. , Laumann, T. O. , Snyder, A. Z. , Schlaggar, B. L. , & Petersen, S. E. (2014). Methods to detect, characterize, and remove motion artifact in resting state fMRI. NeuroImage, 84, 320–341. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Qin, L. , Xu, Q. , Li, Z. , Chen, L. , Li, Y. , Yang, N. , Liu, Z. , Guo, J. , Shen, L. , Allen, E. G. , Chen, C. , Ma, C. , Wu, H. , Zhu, X. , Jin, P. , & Tang, B. (2020). Ethnicity‐specific and overlapping alterations of brain hydroxymethylome in Alzheimer's disease. Human Molecular Genetics, 29, 149–158. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Qiu, T. , Luo, X. , Shen, Z. , Huang, P. , Xu, X. , Zhou, J. , Zhang, M. , & Alzheimer's Disease Neuroimaging Initiative . (2016). Disrupted brain network in progressive mild cognitive impairment measured by eigenvector centrality mapping is linked to cognition and cerebrospinal fluid biomarkers. Journal of Alzheimer's Disease, 54, 1483–1493. [DOI] [PubMed] [Google Scholar]
  59. Quental, N. B. , Brucki, S. M. , & Bueno, O. F. (2013). Visuospatial function in early Alzheimer's disease—The use of the visual object and space perception (VOSP) battery. PLoS One, 8, e68398. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Rathore, S. , Habes, M. , Iftikhar, M. A. , Shacklett, A. , & Davatzikos, C. (2017). A review on neuroimaging‐based classification studies and associated feature extraction methods for Alzheimer's disease and its prodromal stages. NeuroImage, 155, 530–548. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Santos, O. A. , Pedraza, O. , Lucas, J. A. , Duara, R. , Greig‐Custo, M. T. , Hanna Al‐Shaikh, F. S. , Liesinger, A. M. , Bieniek, K. F. , Hinkle, K. M. , Lesser, E. R. , Crook, J. E. , Carrasquillo, M. M. , Ross, O. A. , Ertekin‐Taner, N. , Graff‐Radford, N. R. , Dickson, D. W. , & Murray, M. E. (2019). Ethnoracial differences in Alzheimer's disease from the FLorida Autopsied Multi‐Ethnic (FLAME) cohort. Alzheimer's & Dementia, 15, 635–643. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Schouten, T. M. , Koini, M. , de Vos, F. , Seiler, S. , van der Grond, J. , Lechner, A. , Hafkemeijer, A. , Möller, C. , Schmidt, R. , de Rooij, M. , & Rombouts, S. A. R. B. (2016). Combining anatomical, diffusion, and resting state functional magnetic resonance imaging for individual classification of mild and moderate Alzheimer's disease. Neuroimage: Clinical, 11, 46–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Sendi, M. S. E. , Zendehrouh, E. , Fu, Z. , Liu, J. , Du, Y. , Mormino, E. , Salat, D. , Calhoun, V. , & Miller, R. L. (2021). Disrupted dynamic functional network connectivity among cognitive control networks in the progression of Alzheimer's disease. Brain Connectivity, 13, 334–343. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Sestieri, C. , Shulman, G. L. , & Corbetta, M. (2017). The contribution of the human posterior parietal cortex to episodic memory. Nature Reviews. Neuroscience, 18, 183–192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Smith, S. M. , Jenkinson, M. , Woolrich, M. W. , Beckmann, C. F. , Behrens, T. E. J. , Johansen‐Berg, H. , Bannister, P. R. , De Luca, M. , Drobnjak, I. , Flitney, D. E. , Niazy, R. K. , Saunders, J. , Vickers, J. , Zhang, Y. , De Stefano, N. , Brady, J. M. , & Matthews, P. M. (2004). Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage, 23, S208–S219. [DOI] [PubMed] [Google Scholar]
  66. Sperling, R. A. , Bates, J. F. , Chua, E. F. , Cocchiarella, A. J. , Rentz, D. M. , Rosen, B. R. , Schacter, D. L. , & Albert, M. S. (2003). fMRI studies of associative encoding in young and elderly controls and mild Alzheimer's disease. Journal of Neurology, Neurosurgery, and Psychiatry, 74, 44–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Sperling, R. A. , Jack, C. R., Jr. , & Aisen, P. S. (2011). Testing the right target and right drug at the right stage. Science Translational Medicine, 3, 111cm133. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Tan, C. C. , Yu, J. T. , & Tan, L. (2014). Biomarkers for preclinical Alzheimer's disease. Journal of Alzheimer's Disease, 42, 1051–1069. [DOI] [PubMed] [Google Scholar]
  69. van den Heuvel, M. P. , & Hulshoff Pol, H. E. (2010). Exploring the brain network: A review on resting‐state fMRI functional connectivity. European Neuropsychopharmacology, 20, 519–534. [DOI] [PubMed] [Google Scholar]
  70. van Oostveen, W. M. , & de Lange, E. C. M. (2021). Imaging techniques in Alzheimer's disease: A review of applications in early diagnosis and longitudinal monitoring. International Journal of Molecular Sciences, 22, 2110. [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Wells, R. E. , Yeh, G. Y. , Kerr, C. E. , Wolkin, J. , Davis, R. B. , Tan, Y. , Spaeth, R. , Wall, R. B. , Walsh, J. , Kaptchuk, T. J. , Press, D. , Phillips, R. S. , & Kong, J. (2013). Meditation's impact on default mode network and hippocampus in mild cognitive impairment: A pilot study. Neuroscience Letters, 556, 15–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Xia, M. , Wang, J. , & He, Y. (2013). BrainNet viewer: A network visualization tool for human brain connectomics. PLoS One, 8, e68910. [DOI] [PMC free article] [PubMed] [Google Scholar]
  73. Yamasaki, T. , Horie, S. , Muranaka, H. , Kaseda, Y. , Mimori, Y. , & Tobimatsu, S. (2012). Relevance of in vivo neurophysiological biomarkers for mild cognitive impairment and Alzheimer's disease. Journal of Alzheimer's Disease, 31(Suppl 3), S137–S154. [DOI] [PubMed] [Google Scholar]
  74. Yan, C. G. , Wang, X. D. , Zuo, X. N. , & Zang, Y. F. (2016). DPABI: Data processing & analysis for (resting‐state) brain imaging. Neuroinformatics, 14, 339–351. [DOI] [PubMed] [Google Scholar]
  75. Zang, Y. F. , He, Y. , Zhu, C. Z. , Cao, Q. J. , Sui, M. Q. , Liang, M. , Tian, L. X. , Jiang, T. Z. , & Wang, Y. F. (2007). Altered baseline brain activity in children with ADHD revealed by resting‐state functional MRI. Brain & Development, 29, 83–91. [DOI] [PubMed] [Google Scholar]
  76. Zhao, A. , Li, Y. , Yan, Y. , Qiu, Y. , Li, B. , Xu, W. , Wang, Y. , Liu, J. , & Deng, Y. (2020). Increased prediction value of biomarker combinations for the conversion of mild cognitive impairment to Alzheimer's dementia. Translational Neurodegeneration, 9, 30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  77. Ziontz, J. , Adams, J. N. , Harrison, T. M. , Baker, S. L. , & Jagust, W. J. (2021). Hippocampal connectivity with retrosplenial cortex is linked to neocortical tau accumulation and memory function. The Journal of Neuroscience, 41, 8839–8847. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Data S1. Supporting Information.

Click here for additional data file. (324KB, docx)

Data Availability Statement

The part of datasets supporting the conclusions of this article is available in Alzheimer's Disease Neuroimaging Initiative (ADNI) data repository, http://adni.loni.usc.edu. The Freesurfer version 6.0.0 can be freely downloaded from http://surfer.nmr.mgh.harvard.edu/. DPABI is freely available at http://rfmri.org/dpabi. Group ICA of fMRI Toolbox (GIFT v3.0b) is freely available at http://mialab.mrn.org/software/gift. The BrainNet Viewer is freely available on the NITRC website: http://www.nitrc.org/projects/bnv.

Articles from Human Brain Mapping are provided here courtesy of Wiley

RESOURCES