BrainAGE Estimation: Influence of Field Strength, Voxel Size, Race, and Ethnicity
=================================================================================

* Desarae A. Dempsey
* Rachael Deardorff
* Yu-Chien Wu
* Meichen Yu
* Liana G. Apostolova
* Jared Brosch
* David G. Clark
* Martin R. Farlow
* Sujuan Gao
* Sophia Wang
* Andrew J. Saykin
* Shannon L. Risacher
* the Alzheimer’s Disease Neuroimaging Initiative

## Abstract

The BrainAGE method is used to estimate biological brain age using structural neuroimaging. However, the stability of the model across different scan parameters and races/ethnicities has not been thoroughly investigated. Estimated brain age was compared within- and across-MRI field strength and across voxel sizes. Estimated brain age gap (BAG) was compared across demographically matched groups of different self-reported races and ethnicities in ADNI and IMAS cohorts. Longitudinal ComBat was used to correct for potential scanner effects. The brain age method was stable within field strength, but less stable across different field strengths. The method was stable across voxel sizes. There was a significant difference in BAG between races, but not ethnicities. Correction procedures are suggested to eliminate variation across scanner field strength while maintaining accurate brain age estimation. Further studies are warranted to determine the factors contributing to racial differences in BAG.

Keywords
*   BrainAGE
*   age estimation
*   MRI
*   Alzheimer’s Disease (AD)
*   Mild Cognitive Impairment (MCI)
*   biomarker
*   neurodegeneration
*   brain atrophy
*   cognitive decline
*   race
*   ethnicity
*   ComBat

## 1. Introduction

Brain age estimation techniques have become popular over the last decade as potential biomarkers of individualized brain health and age-related disease severity or prediction [1-5]. The brain age algorithm uses a machine learning model, typically in the form of a regression, trained on structural magnetic resonance imaging (MRI) scans from a large group of healthy individuals from a wide age range to generate a typical pattern of brain aging [1]. The algorithm can then be applied on independent datasets to predict the brain age of individuals, and the deviation from their chronological age can be calculated, which is referred to as the brain age gap (BAG) [1].

Brain aging measures are specifically relevant in patients with Alzheimer’s disease and related dementias (ADRD) due to accelerated brain atrophy in these disorders and the relationship between brain atrophy and cognitive decline [6-11]. The brain age technique has been previously used to measure brain aging in cognitively normal (CN), mild cognitive impairment (MCI), and ADRD [7, 8, 12-19]. Results show that brain age techniques are a potential biomarker for detecting neuroanatomical brain changes that could facilitate early detection before the presentation of clinical AD symptoms [5, 8].

Previous studies have uncovered some biases and limitations in the brain age method. Studies have shown that the brain age framework inherently underestimates the age of older individuals and overestimates the age of younger individuals, compared to their chronological age; corrected procedures are suggested [3, 20-27]. Studies have also provided evidence that the brain age model is stable within- and across-scanner field strength, reported using intraclass correlation coefficient (ICC) values as a measure of test-retest repeatability [2, 3, 12]. However, a linear offset for field strength has been observed [2, 12]. To date, there are no studies comparing stability across different scan voxel sizes, which has been noted as a limitation [3]. Another noted limitation of the brain age model is that majority of studies are conducted in primarily or exclusively non-Hispanic White race participants [1], although a few brain age studies have been conducted in East Asian populations [13, 21, 28]. To our knowledge, there are no studies comparing racial or ethnic groups using the brain age algorithm despite previous studies suggesting that evidence of racial, ethnic, or cultural differences are represented in brain structure [1, 29-36].

The overall aim of this study was to define ways to harmonize the brain age algorithm output in large datasets containing scans with different parameters and individuals from different racial and ethnic groups. Thus, we had two main goals. The first was to determine whether the brain age estimations are stable across different scanner/scan parameters, namely field strength and voxel size. To assess field strength, we used participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) with multiple structural MRI scans in both a 1.5T scanner and a 3T scanner and performed a repeatability analysis within- and across-scanner field strength. Then, we assessed structural MRI scans at three different inherent voxel sizes from participants in the Indiana Memory and Aging Study (IMAS) and performed repeatability analysis across the different voxel sizes. To attempt data harmonization across scanner parameters, we investigated the use of ComBat. ComBat harmonization is a tool that was first created to eliminate batch effects for gene expression microarray data [37]. ComBat has been translated into use for both cross-sectional and longitudinal multi-modal neuroimaging data [38-41]. ComBat is an empirical Bayes method that is useful for removing unwanted technical variability across scanners and sites, while preserving biological variability, and is robust for small sample sizes [37, 38]. Our second goal was to investigate differences in BAG between individuals from different racial or ethnic groups. To accomplish this goal, we assessed differences in BAG between demographically matched African American (AA) individuals and non-Hispanic White (NHW) individuals from IMAS, as well as demographically matched AA and NHW, Asian and NHW, and Hispanic White (HW) and NHW individuals from ADNI.

## 2. Methods

The data presented within this article were from participants from both ADNI ([http://adni.loni.usc.edu](http://adni.loni.usc.edu)) and IMAS. These studies are discussed in detail in the Supplemental text. Details of the sub-samples from each cohort that were used for each parameter investigation are described below. All MRI scans used were 3D T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) scans. Individuals and single scans were excluded for failed BrainAGE processing after visual quality control. Outliers of 3 standard deviations above or below the mean BAG were also excluded.

### 2.1. Consent Statement

Informed consent was obtained according to the Declaration of Helsinki.

### 2.2. ADNI – Field Strength Analysis

ADNI is a multi-site longitudinal study designed to improve and standardize neuroimaging biomarkers for studying AD. For more details, see the Supplementary text. For the field strength analyses, ADNI participants from the ADNI-1 phase who had undergone at least two 1.5T MPRAGE scans in the same session and at least two 3T MPRAGE scans in a separate single session were included. The duplicate scans in different scanner types were conducted on average a few weeks apart. AD and MCI were diagnosed using standard criteria [42, 43]. CN participants had no significant memory concerns or performance deficits on cognitive testing. 25 individuals from each diagnostic group (CN, MCI, AD) were randomly selected and included in the analyses, for a total of 75 participants. All participants were NHW, due to uncertainty of brain age model accuracy on other racial and ethnic groups. 3D T1-weighted MPRAGEs were acquired on various scanners collected at over 50 sites. All scans followed the ADNI-1 MPRAGE sequence (see [http://adni.loni.usc.edu](http://adni.loni.usc.edu)). Pre-processed MRI scans were downloaded from ADNI and processed with using the brain age algorithm as described below.

### 2.3. IMAS – Voxel Size Analysis

The Indiana Memory and Aging Study (IMAS) is a longitudinal observational study of older adults at risk for and with clinical AD. For more details on IMAS and diagnostic criteria, see the Supplemental text and references [42-47]. Included individuals had at least two MPRAGE scans of different inherent voxel sizes collected within the same scan session. Participants had multiple 3D T1-weighted MPRAGEs acquired on a Siemens Prisma 3T scanner at 3 varying inherent voxel sizes: 1) 0.8mm3, 2) 1mm3, 3) 1.1mm x 1.1mm x 1.2mm (termed “ADNI2 sequence”). Specifically, 48 NHW participants (19 CN, 16 SCD, 10 MCI, 3 AD) were included in the analyses, including 47 participants with one 0.8mm3 scan and one ADNI2 sequence scan, and 16 participants with all three scans (one 0.8mm3, one 1mm3, one ADNI2 sequence). Notably, we also tested whether preregistering the scans to Montreal Neurologic Institute (MNI) space affected the brain age results compared to native space scans, which it did not (ICC=0.999 (p<0.001); R2 =0.999 (p<0.001); *data not shown*).

### 2.4. Race/Ethnicity Comparisons (ADNI and IMAS)

In IMAS, 71 AA individuals were each matched as closely as possible on sex, age, education, diagnostic group, and handedness to a NHW individual. All IMAS scans were performed on the same Siemens Prisma 3T scanner and processed using the BrainAGE algorithm described below. In ADNI, 147 AA, 76 HW, and 47 Asian participants were each matched as closely as possible on sex, age, education, diagnostic group, magnet field strength, and handedness to a NHW participant from the same phase of ADNI (ADNI1, ADNIGO/2, ADNI3). Pre-processed 3D MPRAGE scans were downloaded from ADNI ([http://adni.loni.usc.edu](http://adni.loni.usc.edu)) and processed using the BrainAGE algorithm as described below.

### 2.5. Brain Age Framework

Predicted raw brain age estimations were generated using brainageR, (version 1.0, [https://github.com/james-cole/brainageR/releases/tag/1.0](https://github.com/james-cole/brainageR/releases/tag/1.0)), which is performed in R and uses SPM12 and KernLab [3, 14, 48, 49]. A machine-learning Gaussian Process Regression (GPR) was used, which was based on a training dataset of structural MRI scans of 2001 healthy individuals (male/female: 1016/985, mean age: 36.95 (18.12), age range: 18-90 years), to capture the typical brain aging trajectory over time [3, 14]. This GPR was created from multiple public access data sets using different collection sites, various scanners including both 1.5T and 3T, and various voxel sizes [3, 14]. Each participant’s T1-weighted MRI scan from our data set was then registered to standard space, segmented, and compared to this GPR. Then, the code provided an estimated brain age for each participant based on his/her/their brain structure. The BAG is then calculated by subtracting an individual’s chronological age from his/her/their brain age estimation to measure the residual or distance from the GPR, which is a measure of the individual’s deviation from the typical brain aging trajectory. A positive BAG indicates accelerated aging, while a negative BAG indicates decelerated aging compared to the typical brain aging trajectory [2]. For example, an individual that is 67 years old chronologically, but has a brain resembles that of a typical 70-year-old, will have a raw brain age estimation of 70, and a BAG of +3 years, indicating accelerated aging [2].

### 2.6. ComBat Harmonization Method

We used longitudinal ComBat to remove the technical variability seen across scanner field strengths, while maintaining the biological variability in the brain age estimation. Longitudinal ComBat was designed to correct for additive and multiplicative scanner effects while accounting for the within-subject correlation inherent to longitudinal data [38]. Longitudinal ComBat was performed on the brain age values from the repeat 1.5T scans and the repeat 3T scans. The R code for longitudinal ComBat can be found at [https://github.com/jcbeer/longCombat](https://github.com/jcbeer/longCombat) [38].

### 2.7. Statistical Analyses

Statistical analyses were performed using R studio / R version 4.2.1. The raw brain age estimations and BAG scores are reported as mean ± standard deviation in years.

For field strength and voxel-size analyses, intraclass correlation coefficients (ICC: two-way random-effects, absolute agreement, single rater/measurements) and regressions reporting R-squared (R2) values were calculated to assess the repeatability of the brain age output within- and across-scans of different field strengths and across scans of different voxel sizes [3, 45]. In addition, paired t-tests were calculated to determine whether mean brain age estimates were significantly different between scans with different scanner parameters. For the within field strength analyses, a p-value < 0.05 is considered significant. For the across field strength analyses, if Bonferroni correction is applied, a p-value < 0.0125 is considered significant. All p-values far surpassed this cut off for multiple comparison testing, so we report the raw p-values as they are all significant after correction. If Bonferroni correction is applied for the voxel size analyses, a p-value < 0.017 is considered significant.

For the race/ethnicity analyses, independent group t-tests (two-tailed) were used to compare the mean BAG between self-reported racial or ethnic groups. To check appropriate demographic matching, demographic variables were compared between racial groups using a Chi-Square test for nominal variables (diagnostic group, sex, handedness) or an independent t-test (for continuous variables (years of education, scan age)). For the race/ethnicity analyses, no correction is necessary for the independent analysis comparison of AAs and NHWs in the IMAS cohort, and a p-value < 0.05 is considered significant. For the ADNI cohort, each comparison may be considered independent, but since there is potential overlap between the NHW groups, we acknowledge that if Bonferroni correction is applied, a p-value < 0.017 is considered significant.

## 3. Results

### 3.1. Comparisons of Estimated Brain Age within Field Strength

Repeatability scores were consistently high between scans within the same scanner parameters. Raw brain age estimations from two 1.5T scans showed good consistency with ICC=0.992 and R2=0.984 (both p<0.001; Figure 1A, Table 1). Similarly, raw brain age estimations from two 3T scans were consistent with an ICC of 0.987 and R2 =0.975 (both p<0.001; Figure 1B, Table 1). The 1.5T scans gave mean brain age estimates of 74.62 ± 9.69 years and 74.74 ± 9.94 years, while the 3T scans gave mean brain age estimates of 70.49 ± 9.87 years and 70.02 ± 9.45 (Supplemental Table 1). No significant difference within field strength was observed for either the 1.5T or 3T scans (p>0.05; Table 1).

View this table:
[Table 1.](http://medrxiv.org/content/early/2023/12/05/2023.12.05.23299222/T1)

Table 1. 
Comparisons Within and Across Scanner Field Strength in ADNI ICC, R2 values, and paired T-test results for the brain age results are shown for each comparison. There was no significant difference within scanner field strength, but there was a significant difference between every pair across scanner field strength. After ComBat and after linear transformation, there was no longer a significant difference between the brain age values across field strengths. ICC = intraclass correlation coefficient.

![Figure 1.](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/12/05/2023.12.05.23299222/F1.medium.gif)

[Figure 1.](http://medrxiv.org/content/early/2023/12/05/2023.12.05.23299222/F1)

Figure 1. 
Brain Age Estimation Within Field Strength Scatterplots showing the association of raw brain age estimation values **(A)** between scans 1 and 2 within the 1.5T field strength showed an ICC=0.992 (p< 0.001) and R2=0.984 (p<0.001), with the linear equation: y=1.01x-0.35. **(B)** The association between scan 1 and 2 within the 3T field strength shows an ICC=0.987 (p<0.001) and R2=0.975 (p<0.001), with the linear equation: y=0.99x+0.73. ICC = intraclass correlation coefficient

### 3.2 Comparisons of Brain Age Estimates across Field Strength

When comparing across scans from different MRI field strength (1.5T vs 3T), the brain age estimation tool was less consistent than within field strength. There were two 1.5T scans and two 3T scans for each participant, thus there were four possible comparisons: 1.5T scan 1 vs. 3T scan 1, 1.5T scan 1 vs 3T scan 2, 1.5T scan 2 vs 3T scan 1, and 1.5T scan 2 vs 3T scan 2. The ICCs for these comparisons ranged from 0.840 – 0.870 (all p<0.01, Table 1), while R2 values for these comparisons ranged from 0.859 – 0.888 (all p<0.001; Figure 2, Table 1). Linear equations and statistical results describing the brain age estimate relationship between 1.5T and 3T scans are shown in Table 1. Mean brain age estimates for each scan type are shown in Supplemental Table S1 and ranged from 70.02 ± 9.45 for 3T scan 2 to 74.62 ± 9.69 for 1.5T scan 1. Brain age estimates were significantly different for all four comparisons (p<0.001; Table 2). The 3T scanner consistently estimated a lower brain age than the 1.5T scanner, with an offset of 4.43 years on average.

View this table:
[Table 2.](http://medrxiv.org/content/early/2023/12/05/2023.12.05.23299222/T2)

Table 2. Comparisons Across Scanner Voxel Size in IMAS

![Figure 2.](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/12/05/2023.12.05.23299222/F2.medium.gif)

[Figure 2.](http://medrxiv.org/content/early/2023/12/05/2023.12.05.23299222/F2)

Figure 2. 
Brain Age Estimation Across Field Strength Scatterplots showing the correlation of raw brain age estimation values **(A)** between 1.5T scan 1 and 3T scan 1, resulting in repeatability values of: ICC=0.865 (p<0.01) and R2=0.886 (p<0.001), with the linear equation: y=0.96x-1.11 **(B)** between 1.5T scan 1 and 3T scan 2, resulting in repeatability values of: ICC=0.840 (p<0.01) and R2=0.859 (p<0.001), with the linear equation: y=0.95x-0.49 **(C)** between 1.5T scan 2 and 3T scan 1, resulting in repeatability values of: ICC=0.870 (p<0.01) and R2=0.888 (p<0.001), with the linear equation: y=0.94x+0.05 **(D)** between 1.5T scan 2 and 3T scan 2, resulting in repeatability values of: ICC=0.848 (p<0.01) and R2=0.865 (p<0.001), with the linear equation: y=0.94x+0.48. ICC = intraclass correlation coefficient

### 3.3. Comparisons of Brain Age Estimates across Voxel Size

Similar to the field strength analyses above, three different comparisons were calculated to assess differences in raw brain age estimation by voxel size: 0.8mm3 scan vs. 1mm3 scan, 0.8mm3 scan vs. ADNI2 scan (1.1x1.1x1.2mm), and 1mm3 scan vs. ADNI2 scan. The ICC value for the comparison of the 0.8mm3 scan to the 1mm3 scan was 0.968 and the R2 value was 0.965 (all p<0.001; Figure 3A, Table 2). The ICC value for the comparison of the 0.8mm3 scan to the ADNI2 scan was 0.929 and the R2 value was 0.865 (all p<0.001; Figure 3B, Table 2). Finally, the comparison between the 1mm3 scan and the ADNI2 scan had an ICC of 0.965 and R2 of 0.957 (both p<0.001; Figure 3C, Table 2), The mean brain age estimation for 0.8mm3 scan was 70.69 ± 11.05 years, for the 1mm3 scan was 71.64 ± 8.99 years, and for the ADNI2 scan was 71.13 ± 10.45 years (Supplementary Table S2). No significant differences were observed between the mean brain age estimates between the 0.8mm3 and the 1mm3 scans, or the 0.8mm3 and the ADNI2 scans (p>0.05; Table 2). However, a trend for a difference between 1mm3 and ADNI2 scans was observed but did not meet corrected significance (p=0.04; Table 2).

![Figure 3.](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/12/05/2023.12.05.23299222/F3.medium.gif)

[Figure 3.](http://medrxiv.org/content/early/2023/12/05/2023.12.05.23299222/F3)

Figure 3. 
Brain Age Estimation Across Voxel Size Scatterplots showing the correlation of the raw brain age estimation values **(A)** between the 0.8mm3 scan and the 1mm3 scan, resulting in repeatability values of: ICC=0.968 (p<0.001) and R2=0.965 (p<0.001), with the linear equation: y=0.83x+11.94 **(B)** between the 0.8mm3 scan and the ADNI2 scan, resulting in repeatability values of: ICC=0.929 (p<0.001) and R2=0.865 (p<0.001), with the linear equation: y=0.90x+7.66 **(C)** between the 1mm3 scan compared to the ADNI2 sequence (1.1x1.1x1.2mm), resulting in repeatability values of: ICC=0.965 (p<0.001) and R2=0.957 (p<0.001), with the linear equation: y=1.10x-5.78. ICC = intraclass correlation coefficient

### 3.4. ComBat Harmonization Results

Additive effects but no multiplicative effects were found, so longitudinal ComBat was applied for correction. After longitudinal ComBat, there was no longer an additive effect nor significant differences in brain age estimations across field strengths (p>0.05, Table 1). Using the ComBat batch-corrected values, the ICC values increased, while the R2 values slightly decreased. The details are shown in Table 1.

### 3.5 Linear Transformation Results

A linear transformation equation was used to remove differences across scanner field strengths. This linear equation was created by pooling all brain age values from the 1.5T scans and regressing these against all the 3T scans. The regression equation produced was y = 0.95x –0.35. We then applied this equation (adjusted 3T estimate = (raw 3T estimate + 0.35) / 0.95) to all 3T scan brain age values to correct for the scanner difference. After application of this linear transformation, no significant differences across scanner field strength were observed. Further, the ICC values increased, while the R2 values remained the same relative to the original analyses. The details are shown in Table 1.

### 3.6. Comparisons of Brain Age Gap between Demographically Matched Racial and Ethnic Groups

To test BAG between racial and ethnic groups, we first calculated each participant’s BAG (brain age estimate - chronological age). We then compared the mean BAG of each racial or ethnic group (AA, HW, or Asian) to a sample of NHWs matched on diagnostic group, sex, age, education, and handedness, as well as ADNI phase for the ADNI analyses. No significant differences in any demographic variables were detected between racial or ethnic groups, which confirmed appropriate demographic matching (Supplementary Tables S3-S6).

In IMAS, mean BAG was significantly different between NHWs (-3.87 ± 6.79 years) and AA (- 6.49 ± 8.50 years) (t=2.03, p=0.04; Figure 4A, Supplemental Table S3). A similar effect was seen in the ADNI cohort, with a significant difference in mean BAG between the NHW group (- 3.87 ± 8.84 years) and the AA group (-6.79 ± 8.01 years) observed (t=2.96, p=0.003; Figure 4B, Supplemental Table S4). However, BAG was not significantly different between the NHW group (-2.76 ± 8.04 years) and the HW group (-3.75 ± 7.64 years) (t=0.77, p=0.441; Figure 4C, Supplemental Table S5). Finally, a significant difference in BAG between the NHW group (- 1.56 ± 9.64 years) and the Asian group (-5.59 ± 7.80 years) was observed for the raw p-value (t=2.22, p=0.029; Figure 4D, Supplemental Table S6), but not after correction.

![Figure 4.](http://medrxiv.org/http://medrxiv.stage.highwire.org/content/medrxiv/early/2023/12/05/2023.12.05.23299222/F4.medium.gif)

[Figure 4.](http://medrxiv.org/content/early/2023/12/05/2023.12.05.23299222/F4)

Figure 4. 
Brain Age Gap (BAG) Between Race and Ethnicity Groups Box plots comparing BAG between racial and ethnic groups matched on diagnostic group, sex, age, and handedness. **(A)** In the IMAS cohort, a significant difference in BAG between the Non-Hispanic White (NHW) group and the African American (AA) group was observed (t=2.03, p=0.04). **(B)** In the ADNI cohort, there was a significant difference in BAG between NHW and AAs (t=2.96, p=0.003) but **(C)** no difference in BAG between NHWs and Hispanic Whites (HWs) (t=0.77, p=0.441). **(D)** Finally, in ADNI, a significant difference in BAG was observed between NHWs and Asians (t=2.22, p=0.03), which did not remain after correction.

## 4. Discussion

The BrainAGE method was consistent across multiple scans of the same individual in the same scanner, but relatively less stable across scans of the same individual in different scanners with different field strengths. Brain age estimates from scans acquired in a 3T scanner estimated individuals to be ∼4.43 years younger than the estimates from scans acquired on a 1.5T scanner. This finding aligns with previous studies, where stability was lower across field strengths, and it was suggested that researchers correct for the field strength effect by shifting the BAG scores to a zero group mean [3, 12]. Longitudinal ComBat performed relatively well for the correction of scanner field strength. Using ComBat, the need for covarying for scanner is no longer necessary. In addition to the ComBat method, we propose a linear transformation (adjusted 3T estimate = (raw 3T estimate + 0.35) / 0.95), as this equation may be a better fit for correcting for scanner field strength, as it produced higher repeatability values in this study. In addition to supporting the differences by field strength, our results show for the first time that the brain age method was stable across scans of different voxel sizes. However, our group of participants with 1mm3 scans was rather limited (n=16), and thus, this analysis should be repeated in a larger sample, as well as with scans from other inherent voxel sizes.

To our knowledge, this paper is also the first to investigate differences in the brain age method by race or ethnicity. Our results show significant differences in BAG between racial groups (NHWs vs. AA, NHWs vs. Asians), but not between ethnic groups (NHWs vs. Hispanic Whites). However, it is noted that the Asian group did not have a significantly different BAG from NHWs if multiple comparison correction is applied. Both the AA group and the Asian group had lower BAG scores than respective demographically matched NHWs, suggesting decelerated aging in these groups. Previous studies have suggested that there are biological brain differences between East Asian and White, as well as AA and White CN individuals [29-33]. Studies have also suggested differences in presentation of AD and MCI with regard to both cognition and affected brain structures across NHWs, HWs, and AAs [34-36]. However, in the present study, we cannot determine whether the observed differences are biological or a bias from the machine learning model as most of the training set were NHWs. Additional studies should be conducted using larger multi-racial and multi-ethnic cohorts to clearly understand the different results across races from brain age techniques. Further, studies using the brain age technique in samples with mixed races should consider the impact of these factors when performing brain age analyses. Finally, we propose that separate machine learning models should be created and trained on subset of racial groups to ensure accuracy of brain age estimation. ADNI4 will be a prospective cohort that can be used to investigate racial differences further, as it will include a higher percentage of participants from underrepresented backgrounds.

One limitation of our study is the relatively small subgroup sample sizes, especially for the 1mm3 voxel size group (n=16). In addition, all cohorts have a high mean education level, which has been associated with lower brain age results [50]. We did not account for physical activity, which has previously been associated with lower brain age results as well [50]. Finally, while the basis for these findings is still unclear, the observed differences in BAG by race raises several important questions for future research on the neuroimaging methodologies in racial and ethnic groups and highlights the important need for improved recruitment of underrepresented minorities in AD research to better understand aging and age-related diseases in all individuals.

In conclusion, the brain age method was stable within field strength and across inherent voxel sizes. Brain age estimations are dependent on field strength, so we suggest using an offset equation or ComBat to harmonize data in studies with scans from scanners of different field strengths. Lastly, we found differences in BAG scores between self-reported races but are uncertain whether these represent biological differences or bias inherent in the algorithm. Further studies should explore this important area in larger multi-racial samples recruited from population-based studies.

## Supporting information

Supplemental [[supplements/299222_file02.docx]](pending:yes)

## Data Availability

All data produced in the present study are available upon reasonable request.

## Funding

Funding for this project comes from T32 AG071444, P30 AG010133, P30 AG072976, R01 AG019771, R01 AG057739, U19 AG024904, R01 LM013463, R01 AG068193, U01 AG068057, and U01 AG072177.

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](http://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.

## Collaborators

*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](http://adni.loni.usc.edu/wp-content/uploads/how\_to_apply/ADNI_Acknowledgement_List.pdf)

## Disclosures

Dr. Apostolova received grant or other financial support from the National Institutes of Health (NIH), Alzheimer’s Association, AVID Pharmaceuticals, Life Molecular Imaging, Roche Diagnostics, and Eli Lilly. In addition, she has received consulting fees from Biogen, Two Labs, IQIVA, Florida Department of Health, Genentech, NIH Biobank, Eli Lilly, GE Healthcare, Eisai, and Roche Diagnostics. She has also received payment or honoraria from American Academy of Neurology, MillerMed, National Alzheimer’s Coordinating Center CME, CME Institute, APhA, Purdue University, Mayo Clinic, MJH Physician Education Resource, and Ohio State University. She received support for travel from the Alzheimer’s Association. She has served on Data Safety and Monitoring or Advisory Boards for IQVIA, UAB Nathan Schock Center, New Mexico Exploratory ADRC, and NIA R01 AG061111. She has a leadership role in multiple committees, including the Medical Science Council of the Alzheimer’s Association Greater Indiana Chapter, the Alzheimer’s Association Science Program Committee, and the FDA PCNS Advisory Committee. Finally, Dr. Apostolova holds stock in Cassava Neurosciences and Golden Seeds. Dr. Saykin receives support from multiple NIH grants (P30 AG010133, P30 AG072976, R01 AG019771, R01 AG057739, U19 AG024904, R01 LM013463, R01 AG068193, T32 AG071444, U01 AG068057, U01 AG072177, and U19 AG074879). He has also received support from Avid Radiopharmaceuticals, a subsidiary of Eli Lilly (in kind contribution of PET tracer precursor); Bayer Oncology (Scientific Advisory Board); Eisai (Scientific Advisory Board); Siemens Medical Solutions USA, Inc. (Dementia Advisory Board); NIH NHLBI (MESA Observational Study Monitoring Board); Springer-Nature Publishing (Editorial Office Support as Editor-in-Chief, Brain Imaging and Behavior). All other authors declare no conflict of interest.

## Acknowledgements

The authors thank Steve Brown, Bobi Eastman, Tatiana M. Foroud, Yolanda Graham-Dotson, Tyler Gannon, Savannah Hottle, Gary D. Hutchins, Brenna McDonald, Fredrick Unverzagt, Aaron Vosmeier, and Donna Wert for their contributions to this work.

*   Received December 5, 2023.
*   Revision received December 5, 2023.
*   Accepted December 5, 2023.


*   © 2023, Posted by Cold Spring Harbor Laboratory

This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/)

## References

1.  [1].Baecker L, Garcia-Dias R, Vieira S, Scarpazza C, Mechelli A. Machine learning for brain age prediction: Introduction to methods and clinical applications. EBioMedicine. 2021;72:103600. doi:10.1016/j.ebiom.2021.103600
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ebiom.2021.103600&link_type=DOI) 

2.  [2].Franke K, Gaser C. Ten Years of BrainAGE as a Neuroimaging Biomarker of Brain Aging: What Insights Have We Gained? Front Neurol. 2019;10:789. doi:10.3389/fneur.2019.00789
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.3389/fneur.2019.00789&link_type=DOI) 

3.  [3].Cole JH, Poudel RPK, Tsagkrasoulis D, Caan MWA, Steves C, Spector TD, et al. Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker. Neuroimage. 2017;163:115–24. doi:10.1016/j.neuroimage.2017.07.059
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2017.07.059&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

4.  [4].Cole JH, Franke K. Predicting Age Using Neuroimaging: Innovative Brain Ageing Biomarkers. Trends Neurosci. 2017;40:681–90. doi:10.1016/j.tins.2017.10.001
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.tins.2017.10.001&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

5.  [5].Elliott ML, Belsky DW, Knodt AR, Ireland D, Melzer TR, Poulton R, et al. Brain-age in midlife is associated with accelerated biological aging and cognitive decline in a longitudinal birth cohort. Mol Psychiatry. 2021;26:3829–38. doi:10.1038/s41380-019-0626-7
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/s41380-019-0626-7&link_type=DOI) 

6.  [6].Fjell AM, Walhovd KB, Fennema-Notestine C, McEvoy LK, Hagler DJ, Holland D, et al. One-year brain atrophy evident in healthy aging. J Neurosci. 2009;29:15223–31. doi:10.1523/JNEUROSCI.3252-09.2009
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Njoiam5ldXJvIjtzOjU6InJlc2lkIjtzOjExOiIyOS80OC8xNTIyMyI7czo0OiJhdG9tIjtzOjUwOiIvbWVkcnhpdi9lYXJseS8yMDIzLzEyLzA1LzIwMjMuMTIuMDUuMjMyOTkyMjIuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

7.  [7].Gaser C, Franke K, Klöppel S, Koutsouleris N, Sauer H, ADNI. BrainAGE in Mild Cognitive Impaired Patients: Predicting the Conversion to Alzheimer’s Disease. PLoS ONE. 2013;8:e67346. doi:10.1371/journal.pone.0067346
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1371/journal.pone.0067346&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23826273&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

8.  [8].Franke K, Ziegler G, Kloppel S, Gaser C, Alzheimer’s Disease Neuroimaging I. Estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: exploring the influence of various parameters. Neuroimage. 2010;50:883–92. doi:10.1016/j.neuroimage.2010.01.005
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2010.01.005&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20070949&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

9.  [9].Davatzikos C, Xu F, An Y, Fan Y, Resnick SM. Longitudinal progression of Alzheimer’s-like patterns of atrophy in normal older adults: the SPARE-AD index. Brain. 2009;132:2026–35. doi:10.1093/brain/awp091
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/brain/awp091&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19416949&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000268330500004&link_type=ISI) 

10. [10].Sluimer JD, van der Flier WM, Karas GB, van Schijndel R, Barnes J, Boyes RG, et al. Accelerating regional atrophy rates in the progression from normal aging to Alzheimer’s disease. Eur Radiol. 2009;19:2826–33. doi:10.1007/s00330-009-1512-5
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1007/s00330-009-1512-5&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19618189&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000271949500005&link_type=ISI) 

11. [11].Liem F, Varoquaux G, Kynast J, Beyer F, Kharabian Masouleh S, Huntenburg JM, et al. Predicting brain-age from multimodal imaging data captures cognitive impairment. Neuroimage. 2017;148:179–88. doi:10.1016/j.neuroimage.2016.11.005
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2016.11.005&link_type=DOI) 

12. [12].Franke K, Gaser C, ADNI. Longitudinal Changes in Individual BrainAGE in Healthy Aging, Mild Cognitive Impairment, and Alzheimer’s Disease. GeroPsych - The Journal of Gerontopsychology and Geriatric Psychiatry. 2012;25:235–45. doi:10.1024/1662-9647/a000074
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1024/1662-9647/a000074&link_type=DOI) 

13. [13].Beheshti I, Maikusa N, Matsuda H. The association between “Brain-Age Score” (BAS) and traditional neuropsychological screening tools in Alzheimer’s disease. Brain Behav. 2018;8:e01020. doi:10.1002/brb3.1020
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/brb3.1020&link_type=DOI) 

14. [14].Cole JH, Ritchie SJ, Bastin ME, Valdes Hernandez MC, Munoz Maniega S, Royle N, et al. Brain age predicts mortality. Mol Psychiatry. 2018;23:1385–92. doi:10.1038/mp.2017.62
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/mp.2017.62&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=28439103&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

15. [15].Ly M, Yu GZ, Karim HT, Muppidi NR, Mizuno A, Klunk WE, et al. Improving brain age prediction models: incorporation of amyloid status in Alzheimer’s disease. Neurobiol Aging. 2020;87:44–8. doi:10.1016/j.neurobiolaging.2019.11.005
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neurobiolaging.2019.11.005&link_type=DOI) 

16. [16].Lowe LC, Gaser C, Franke K, Alzheimer’s Disease Neuroimaging I. The Effect of the APOE Genotype on Individual BrainAGE in Normal Aging, Mild Cognitive Impairment, and Alzheimer’s Disease. PLoS One. 2016;11:e0157514. doi:10.1371/journal.pone.0157514
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1371/journal.pone.0157514&link_type=DOI) 

17. [17].Kaufmann T, van der Meer D, Doan NT, Schwarz E, Lund MJ, Agartz I, et al. Common brain disorders are associated with heritable patterns of apparent aging of the brain. Nat Neurosci. 2019;22:1617–23. doi:10.1038/s41593-019-0471-7
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/s41593-019-0471-7&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=31551603&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

18. [18].Mohajer B, Abbasi N, Mohammadi E, Khazaie H, Osorio RS, Rosenzweig I, et al. Gray matter volume and estimated brain age gap are not linked with sleep-disordered breathing. Hum Brain Mapp. 2020;41:3034–44. doi:10.1002/hbm.24995
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/hbm.24995&link_type=DOI) 

19. [19].Varikuti DP, Genon S, Sotiras A, Schwender H, Hoffstaedter F, Patil KR, et al. Evaluation of non-negative matrix factorization of grey matter in age prediction. Neuroimage. 2018;173:394–410. doi:10.1016/j.neuroimage.2018.03.007
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2018.03.007&link_type=DOI) 

20. [20].Smith SM, Vidaurre D, Alfaro-Almagro F, Nichols TE, Miller KL. Estimation of brain age delta from brain imaging. Neuroimage. 2019;200:528–39. doi:10.1016/j.neuroimage.2019.06.017
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2019.06.017&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=31201988&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

21. [21].Aycheh HM, Seong JK, Shin JH, Na DL, Kang B, Seo SW, et al. Biological Brain Age Prediction Using Cortical Thickness Data: A Large Scale Cohort Study. Front Aging Neurosci. 2018;10:252. doi:10.3389/fnagi.2018.00252
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.3389/fnagi.2018.00252&link_type=DOI) 

22. [22].de Lange AG, Cole JH. Commentary: Correction procedures in brain-age prediction. Neuroimage Clin. 2020;26:102229. doi:10.1016/j.nicl.2020.102229
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.nicl.2020.102229&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=32120292&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

23. [23].Le TT, Kuplicki RT, McKinney BA, Yeh HW, Thompson WK, Paulus MP, et al. A Nonlinear Simulation Framework Supports Adjusting for Age When Analyzing BrainAGE. Front Aging Neurosci. 2018;10:317. doi:10.3389/fnagi.2018.00317
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.3389/fnagi.2018.00317&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30405393&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

24. [24].Liang H, Zhang F, Niu X. Investigating systematic bias in brain age estimation with application to post-traumatic stress disorders. Hum Brain Mapp. 2019;40:3143–52. doi:10.1002/hbm.24588
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/hbm.24588&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30924225&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

25. [25].Niu X, Zhang F, Kounios J, Liang H. Improved prediction of brain age using multimodal neuroimaging data. Hum Brain Mapp. 2020;41:1626–43. doi:10.1002/hbm.24899
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/hbm.24899&link_type=DOI) 

26. [26].Cole JH. Multimodality neuroimaging brain-age in UK biobank: relationship to biomedical, lifestyle, and cognitive factors. Neurobiol Aging. 2020;92:34–42. doi:10.1016/j.neurobiolaging.2020.03.014
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neurobiolaging.2020.03.014&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=32380363&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

27. [27].Treder MS, Shock JP, Stein DJ, du Plessis S, Seedat S, Tsvetanov KA. Correlation Constraints for Regression Models: Controlling Bias in Brain Age Prediction. Front Psychiatry. 2021;12:615754. doi:10.3389/fpsyt.2021.615754
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.3389/fpsyt.2021.615754&link_type=DOI) 

28. [28].Simfukwe C, Youn YC. Prediction of East Asian Brain Age using Machine Learning Algorithms Trained With Community-based Healthy Brain MRI. Dement Neurocogn Disord. 2022;21:138–46. doi:10.12779/dnd.2022.21.4.138
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.12779/dnd.2022.21.4.138&link_type=DOI) 

29. [29].Isamah N, Faison W, Payne ME, MacFall J, Steffens DC, Beyer JL, et al. Variability in frontotemporal brain structure: the importance of recruitment of African Americans in neuroscience research. PLoS One. 2010;5:e13642. doi:10.1371/journal.pone.0013642
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1371/journal.pone.0013642&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21049028&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

30. [30].Chee MW, Zheng H, Goh JO, Park D, Sutton BP. Brain structure in young and old East Asians and Westerners: comparisons of structural volume and cortical thickness. J Cogn Neurosci. 2011;23:1065–79. doi:10.1162/jocn.2010.21513
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1162/jocn.2010.21513&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20433238&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000289062000004&link_type=ISI) 

31. [31].Kang DW, Wang S-M, Na H-R, Park SY, Kim NY, Lee CU, et al. Differences in cortical structure between cognitively normal East Asian and Caucasian older adults: a surface-based morphometry study. Scientific Reports. 2020;10:20905. doi:10.1038/s41598-020-77848-8
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1038/s41598-020-77848-8&link_type=DOI) 

32. [32].Tang Y, Zhao L, Lou Y, Shi Y, Fang R, Lin X, et al. Brain structure differences between Chinese and Caucasian cohorts: A comprehensive morphometry study. Hum Brain Mapp. 2018;39:2147–55. doi:10.1002/hbm.23994
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/hbm.23994&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=29400417&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

33. [33].Zilles K, Kawashima R, Dabringhaus A, Fukuda H, Schormann T. Hemispheric shape of European and Japanese brains: 3-D MRI analysis of intersubject variability, ethnical, and gender differences. Neuroimage. 2001;13:262–71. doi:10.1006/nimg.2000.0688
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1006/nimg.2000.0688&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=11162267&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000166729100003&link_type=ISI) 

34. [34]. B. Zahodne L,  J. Manly J, Narkhede A,  Y. Griffith E, DeCarli C,  S. Schupf N, et al. Structural MRI Predictors of Late-Life Cognition Differ Across African Americans, Hispanics, and Whites. Current Alzheimer Research. 2015;12:632–9.
    
    

35. [35].DeCarli C, Reed BR, Jagust W, Martinez O, Ortega M, Mungas D. Brain behavior relationships among African Americans, whites, and Hispanics. Alzheimer Dis Assoc Disord. 2008;22:382–91. doi:10.1097/wad.0b013e318185e7fe
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1097/WAD.0b013e318185e7fe&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19068502&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000261420000010&link_type=ISI) 

36. [36].Gavett BE, Fletcher E, Harvey D, Farias ST, Olichney J, Beckett L, et al. Ethnoracial differences in brain structure change and cognitive change. Neuropsychology. 2018;32:529–40. doi:10.1037/neu0000452
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1037/neu0000452&link_type=DOI) 

37. [37].Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics. 2007;8:118–27. doi:10.1093/biostatistics/kxj037
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/biostatistics/kxj037&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=16632515&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000242715400008&link_type=ISI) 

38. [38].Beer JC, Tustison NJ, Cook PA, Davatzikos C, Sheline YI, Shinohara RT, et al. Longitudinal ComBat: A method for harmonizing longitudinal multi-scanner imaging data. Neuroimage. 2020;220:117129. doi:10.1016/j.neuroimage.2020.117129
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2020.117129&link_type=DOI) 

39. [39].Fortin JP, Cullen N, Sheline YI, Taylor WD, Aselcioglu I, Cook PA, et al. Harmonization of cortical thickness measurements across scanners and sites. Neuroimage. 2018;167:104–20. doi:10.1016/j.neuroimage.2017.11.024
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2017.11.024&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=29155184&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

40. [40].Fortin JP, Parker D, Tunc B, Watanabe T, Elliott MA, Ruparel K, et al. Harmonization of multi-site diffusion tensor imaging data. Neuroimage. 2017;161:149–70. doi:10.1016/j.neuroimage.2017.08.047
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neuroimage.2017.08.047&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

41. [41].Yu M, Linn KA, Cook PA, Phillips ML, McInnis M, Fava M, et al. Statistical harmonization corrects site effects in functional connectivity measurements from multi-site fMRI data. Hum Brain Mapp. 2018;39:4213–27. doi:10.1002/hbm.24241
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/hbm.24241&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=29962049&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

42. [42].McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., Kawas CH, et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7:263–9. doi:10.1016/j.jalz.2011.03.005
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jalz.2011.03.005&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21514250&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000291239600003&link_type=ISI) 

43. [43].Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Med. 2004;256:183–94. doi:10.1111/j.1365-2796.2004.01388.x
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1111/j.1365-2796.2004.01388.x&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=15324362&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000223470900003&link_type=ISI) 

44. [44].Jessen F, Amariglio RE, van Boxtel M, Breteler M, Ceccaldi M, Chetelat G, et al. A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement. 2014;10:844–52. doi:10.1016/j.jalz.2014.01.001
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jalz.2014.01.001&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=24798886&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

45. [45].Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15:155–63. doi:10.1016/j.jcm.2016.02.012
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jcm.2016.02.012&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=27330520&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

46. [46].Rattanabannakit C, Risacher SL, Gao S, Lane KA, Brown SA, McDonald BC, et al. The Cognitive Change Index as a Measure of Self and Informant Perception of Cognitive Decline: Relation to Neuropsychological Tests. J Alzheimers Dis. 2016;51:1145–55. doi:10.3233/JAD-150729
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NjoiaW9zamFkIjtzOjU6InJlc2lkIjtzOjk6IjUxLzQvMTE0NSI7czo0OiJhdG9tIjtzOjUwOiIvbWVkcnhpdi9lYXJseS8yMDIzLzEyLzA1LzIwMjMuMTIuMDUuMjMyOTkyMjIuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

47. [47].Weintraub S, Besser L, Dodge HH, Teylan M, Ferris S, Goldstein FC, et al. Version 3 of the Alzheimer Disease Centers’ Neuropsychological Test Battery in the Uniform Data Set (UDS). Alzheimer Dis Assoc Disord. 2018;32:10–7. doi:10.1097/WAD.0000000000000223
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1097/WAD.0000000000000223&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom) 

48. [48].Karatzoglou A, Smola A, Hornik K. kernlab: Kernel-Based Machine Learning Lab. 2022.
    
    

49. [49].Karatzoglou A, Smola A, Hornik K, Zeileis A. kernlab -- An {S4} Package for Kernel Methods in {R}. Journal of Statistical Software. 2004;11:1--20. doi:10.18637/jss.v011.i09
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.18637/jss.v011.i09&link_type=DOI) 

50. [50].Steffener J, Habeck C, O’Shea D, Razlighi Q, Bherer L, Stern Y. Differences between chronological and brain age are related to education and self-reported physical activity. Neurobiol Aging. 2016;40:138–44. doi:10.1016/j.neurobiolaging.2016.01.014
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.neurobiolaging.2016.01.014&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=26973113&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2023%2F12%2F05%2F2023.12.05.23299222.atom)