8. Reproduce Paper Results

We refer the reader/user to https://brainsuite.github.io/bstrPaper for instructions and resources to reproduce the results presented in our paper bstr: The BrainSuite Statistics Toolbox in R (preprint to be available on bioRxiv). These results were derived from data retrieved from two open-access sources: OASIS-3 (LaMontagne et al., 2019) and the NKI Rockland Sample (Nooner et al., 2012). The Rockland source data were used under a CC BY-NC license, and we are thus able to share our processed outputs. These are also released under a CC BY-NC license. The OASIS-3 data require a data use agreement and other conditions (see https://sites.wustl.edu/oasisbrains/home/oasis-3/ for details).

Detailed documentation for bstr is available from https://brainsuite.org/bstr.

The results shown in our paper are intended to provide examples of the capabilities of bstr. These include statistical analyses performed on data processed using BrainSuite’s pipelines for processing anatomical and diffusion data. This processing was performed using the BrainSuite BIDS App (Kim et al., 2024).

We are sharing the following (listed in order of increasing difficulty for the user): * The final bstr outputs for the OASIS3 and Rockland data. These can be visualized in a browser or in the BrainSuite GUI, depending on the result type. * The BrainSuite outputs on the Rockland data. These can be processed with R commands (also provided) to reproduce the group-level analysis of diffusion MRI data presented in Sec. 3.2 of our paper. * Instructions for full reproduction of the results in our paper. This requires retrieving large amounts of data and substantial compute time.

8.1. Visualization of OASIS-3 and Rockland dataset final results (quick)

The final outputs for the analyses shown in our paper are provided within this repository.

You can view these results locally by cloning this repository or by downloading and unpacking a 700MB zipfile of it, and then following the directions below.

We also host the bstr reports online on our bstr Paper Data website.

8.1.1. Visualization of OASIS-3 dataset final results (quick)

  1. To view the ROI results for group differences for CDR_BIN in the Left Superior Temporal Gyrus (STG), go to the folder OASIS-3_bstr_results/roi_anova_CDR_BIN and open report_bstr_lm_CDR_BIN.html.

  2. To view the ROI results for correlation of MMSE with GM thickness in the Right STG, go to the folder OASIS-3_bstr_results/roi_corr_MMSE and open report_bstr_corr_MMSE.html.

  3. To view the SBA results for the main effect of MMSE on cortical thickness, go to the folder OASIS-3_bstr_results/sba_anova_MMSE_smooth2 and drag and drop the left and right hemisphere surfaces bstr_anova_MMSE_BCI-DNI_brain.left.mid.cortex_log_pvalues_adjusted.dfs and bstr_anova_MMSE_BCI-DNI_brain.right.mid.cortex_log_pvalues_adjusted.dfs in the BrainSuite GUI.

  4. To view the SBA results for an unpaired t-test for the CDR_BIN groups, go to the folder OASIS-3_bstr_results/sba_unpaired_ttest_CDR_BIN and drag and drop the left and right hemisphere surfaces unpairedttest_CDR_BIN_BCI-DNI_brain.left.mid.cortex_log_pvalues_adjusted.dfs and unpairedttest_CDR_BIN_BCI-DNI_brain.right.mid.cortex_log_pvalues_adjusted.dfs in the BrainSuite GUI.

  5. To view the TBM results for the correlation of CDR with the magnitue of the volumetric deformation fields, go to the folder OASIS-3_bstr_results/tbm_corr_CDR_smooth3 and open the HTML report report_bstr_corr_CDR.html.

  6. Finally, to view the TBM results for the correlation of Age (Figure 5 in the paper), go to the folder OASIS-3_bstr_results/tbm_corr_Age_smooth3 and open the HTML report report_bstr_corr_Age.html.

8.1.2. Visualization of Rockland dataset final results (quick)

  1. To view the DBA results for age effect on FA, go to the output folder Rockland_bstr_results/dba_anova_age_FA_smooth3 and open report_bstr_anova_age.html.

  2. To view the DBA results for correlation of age with MD, go to the output folder Rockland_bstr_results/dba_corr_age_MD_smooth3 and open report_bstr_corr_age.html.

8.2. Reproducing bstr analysis for the Rockland dataset (~10 mins)

  1. Download the Rockland diffusion data Rockland_BrainSuite_Processed.zip (18GB), which we have processed with BrainSuite.

  2. Download the demographics file Rockland.csv and the R-notebook Rockland.Rmd.

  3. Open Rockland.Rmd in and editor and change the data paths. The data_dir should point to the unzipped Rockland diffusion output folder, the outdir_leading should point to an existing output directory where you want to store the results of bstr analysis, and the test_csv should point to the downloaded Rockland.csv.

  4. Execute bstr analysis using either of the two following steps:

    1. If using RStudio, with the Rockland.Rmd currently open, click on File -> Knit Document.

    2. If using the R console, make sure rmarkdown is installed. If not, type install.packages("rmarkdown") to install it. Then go to the directory (setwd()) where the R-notebook is saved, and type rmarkdown::render("Rockland.Rmd").

    If the paths are set correctly, these commands should execute all the sections of the Rmd and create two output folders in the outdir_leading output directory: dba_anova_age_FA_smooth3 for the FA analysis and dba_corr_age_MD_smooth3 for the MD analysis.

8.3. Reproducing bstr analysis from source data

The analyses described in our paper are fully reproducible from the source data. Performing these steps is compute and storage intensive.

8.3.1. Reproducing bstr analysis for the OASIS-3 dataset

The OASIS-3 dataset is part of the Open Access Series of Imaging Studies (OASIS) and is described in this paper by LaMontagne et al., (2019). The data can be downloaded from here after signing the data use agreement. Because of this agreement, we are unable to redistribute the OASIS-3 data or provide pre-processed inputs to bstr.

Users wishing to fully reproduce the results in the paper can do so by retrieving the data from the OASIS-3 database and processing it with the Anatomical Pipeline in the BrainSuite BIDS App. Please see the BrainSuite BIDS App site for details on how to perform this processing.

Once processed, the data can be analyzed with bstr using the R-notebook OASIS-3.Rmd and the OASIS-3 demographics files OASIS-3.csv.

8.3.2. Reproducing bstr analysis for the Rockland dataset

The the NKI Rockland Sample (Nooner et al., 2012) is available under a CC BY-NC license. As described above, we have made available: (a) the outputs of the bstr analysis described in our paper; and (b) the outputs of the participant-level stage (including the Anatomical Pipeline and Diffusion Pipeline) of the the BrainSuite BIDS App applied to the Rockland data.

Users wishing to fully reproduce the analyses from the source data can download the Rockland data from https://fcon_1000.projects.nitrc.org/indi/pro/nki.html and process each subject with the BrainSuite BIDS App. They can then proceed with the bstr analysis described in the previous section.