In-vivo brain MRI data are collected through the RADC cohort studies and substudies. Participants are imaged biennially using a number of pulse sequences providing structural, functional, and chemical information about the brain. The raw data are processed locally. We use a 3-step quality control strategy testing: 1) MRI scanner performance on a phantom, 2) quality of raw MRI data collected on humans, and 3) quality of information derived from processing. Each step involves thorough tests tailored to each MRI processing output as well as visual inspection. Processing of the raw MR images generates the following output:

1. Freesurfer output: Distortion-corrected MPRAGE data is segmented into Desikan-Killiany cortical and subcortical regions using Freesurfer (Fischl et al., Cereb Cortex 2004;14:11-22) (surfer.nmr.mgh.harvard.edu). All output is visually inspected, corrections are made when necessary and Freesurfer is rerun. Regional volumes and cortical thicknesses are calculated. Freesurfer output is not generated when the MPRAGE data is of low signal to noise ratio (SNR), low gray-white matter contrast to noise ratio (CNR), or is contaminated by motion artifacts. Partial Freesurfer output is generated when most of the brain allows reliable segmentation while some regions cannot be reliably segmented (e.g. due to low contrast or lesions). (Fleischman et al., Brain Struct Funct 2014; 219:2029-2049)

   Regional cortical thickness generated by Freesurfer (THK_ROI_**M): The thicknesses of homologous regions in contralateral hemispheres are averaged. To find the name of the region corresponding to each variable name use the lookup table for the gray matter nodes of the IIT white matter atlas (www.nitrc.org/projects/iit).

   Regional gray matter volumes generated by Freesurfer (VOLG_ROI_**M): The volumes of homologous regions in contralateral hemispheres are averaged. The average volumes are then divided by the intracranial volume and the ratio is finally multiplied by 1000. To find the name of the region corresponding to each variable name use the lookup table for the gray matter nodes of the IIT white matter atlas (www.nitrc.org/projects/iit).

2. Total volumes: Distortion-corrected MPRAGE data is segmented into gray matter, white matter and cerebrospinal fluid (CSF) using the Computational Anatomy Toolbox (CAT) (www.neuro.uni-jena.de/cat/). The total volumes of gray matter, white matter, and CSF are calculated. The intracranial volume is calculated as the sum of the gray matter, white matter and CSF volumes. Total volumes are not calculated when the MPRAGE data is of low signal to noise ratio (SNR), low gray-white matter contrast to noise ratio (CNR), or is contaminated by motion artifacts. (For 3T data, total volumes are calculated as described in this section, while for 1.5T data the total volumes are generated by Freesurfer).

   Gray matter volume (GMPCT) is expressed as percent of the intracranial volume. The volume of the gray matter is divided by the intracranial volume and multiplied by 100.

   White matter volume (WMPCT) is expressed as a percent of the intracranial volume. The volume of the white matter is divided by the intracranial volume and multiplied by 100.

   Cerebrospinal fluid volume (CSFPCT) is expressed as a percent of the intracranial volume. The volume of the cerebrospinal fluid is divided by the intracranial volume and multiplied by 100.

   Intracranial volume (ICV) is measured in mm³.

3. White matter hyperintensities (WMH): White matter lesions appearing hyperintense in T2-weighted images are segmented based on FLAIR and T1-weighted data using sysu (Li H, et al. Neuroimage 2018;183:650-665). A mask of white matter hyperintensities is generated and the total volume of hyperintensities is calculated. White matter hyperintensities are not segmented when the FLAIR data is of low signal to noise ratio (SNR) or is contaminated by motion artifacts. (Lamar et al., Am J Geriatr Psychiatry. 2023;31:1129-1139)

   White matter hyperintensities (WMHPCT_LOG10) volume is the logarithm base 10 of the percent of the intracranial volume occupied by WMH. The total volume of WMH is divided by the intracranial volume, multiplied by 100, and then transformed with the logarithm base 10.

4. Regional gray matter volumes generated by multi-atlas segmentation (MAS_ROI_**M) Gray matter is first segmented and then divided into regions based on multi-atlas segmentation. The volumes of homologous regions in contralateral hemispheres are averaged. The average volumes are then divided by the intracranial volume and the ratio is multiplied by 1000. To find the name of the region corresponding to each variable name use the lookup table for the gray matter nodes of the IIT white matter atlas www.nitrc.org/projects/iit. Note: There are 3 variables which refer to combinations of two regions. These are MAS_ROI_05_15M: entorhinal and parahippocampal; MAS_ROI_26_31M: frontal pole and rostral middle frontal; and MAS_ROI_29_32M: temporal pole and superior temporal.

5. Median R2 values for gray matter regions segmented by multi-atlas segmentation (R2med_ROI_**) Maps of R2 relaxation times are generated by fitting the multi-echo fast spin-echo signals with a mono-exponential decay model. For each gray matter region segmented by the multi-atlas segmentation, the R2 values from all voxels of the region in both the left and right hemisphere are grouped together and their median value is identified. The R2 values have units ms-1. To find the name of the region corresponding to each variable name use the lookup table for the gray matter nodes of the IIT white matter atlas www.nitrc.org/projects/iit. Note 1: There are 3 variables which refer to combinations of two regions. These are MAS_ROI_05_15M: entorhinal and parahippocampal; MAS_ROI_26_31M: frontal pole and rostral middle frontal; and MAS_ROI_29_32M: temporal pole and superior temporal. Note 2: The mean of the medians of all cortical regions and the mean of the medians of all subcortical regions are also provided.

6. ARTS score (ARTS_SCORE). ARTS is a fully automated biomarker that outputs a score linked to the likelihood a person suffers from arteriolosclerosis based on in-vivo brain MRI data and basic demographic information. The higher the score, the higher the likelihood of arteriolosclerosis. For more information, see: www.nitrc.org/projects/arts (Makkinejad et al. Neuroimage Clin. 2021;31:102768)

7. Brain age (AGE_BRAIN_MRI). Brain age is estimated from structural brain MRI data using a deep neural network (Leonardsen et al., Neuroimage 2022;256:119210). The model has been fine-tuned using RADC data from participants that had no cognitive impairment. Brain age is expressed in years.

8. Cortical thickness (AD signature) (AD_SIGNAT_THCK). Cortical thickness for the AD signature is the mean of the thickness (in mm) of the following cortical regions from the two hemispheres: middle temporal and inferior temporal, entorhinal, temporal pole, fusiform, superior frontal, rostral and caudal middle frontal, superior and inferior parietal, supramarginal, and precuneus. When the measurement of the thickness of a region in one of the two hemispheres is rejected during quality checks, then only the thickness of a region in one of the two hemispheres is used. (Lamar et al. Am J Geriatr Psychiatry. 2023;31:1129-1139).

Documentation in progress.