mnf-article-data-analysis

Data and MATLAB code for statistical analysis of MNF acquisitions in brain at 7 T (for MNF article)

CONTACT

guillaume.madelin@nyulangone.org

MATLAB

The code was written and tested with MATLAB 2025b in Windows 11, and requires the Image Processing and the Statistics and Machine Learning toolboxes.

DATA OVERVIEW

Multinuclear fingerprinting (MNF) consists of data acquisition with simultaneous proton (¹H) and sodium (²³Na) magnetic resonance fingerprinting (MRF) [1-5] and data post-processing with a super-resolution (SR) algorithm [6] to generate final ²³Na maps with the same resolution as the ¹H maps.

In summary: MNF = Simultaneous ¹H/²³Na MRF + SR.

MNF maps

• Three ¹H maps: Proton Density (PD), and relaxation times T₁ and T₂
• Four ²³Na maps: Tissue Sodium Concentration (TSC), and relaxation times T₁, T_2s (short T₂ component) and T_2l (long T₂ component)

MNF resolution

• High Resolution (HR): 1.5 x 1.5 x 5 mm³
• Low Resolution (LR): 2.85 x 2.85 x 5 mm³

Regions-of-interest (ROI)

• EYE: Vitreous humor in the eyes
• GM: Grey Matter
• WM: White Matter
• CSF: Cerebrospinal Fluid

FOLDERS

data_251027

This folder contains all the MNF 3D data from 7 subjects in nifti format, including:

• Subjects 1-7 (subfolders s1 to s7):
  • ¹H MNF HR: PD, T₁, T₂
  • ²³Na MNF HR: TSC, T₁, T_2s, T_2l
  • ²³Na MNF LR: TSC, T₁, T_2s, T_2l
  • Binary masks for ROIs: EYE, CSF, GM, WM
  • Probability maps from SPM12: EYE, CSF, GM, WM
• Subjects 5 & 6 (additional data in sub-folders s5 & s6):
  • ²³Na FLORET LR: TSC, T₁, T_2s, T_2l

matlab

This folder contains the Matlab code for analyzing the data in folder "data_251027", including:

• run_analyze_mnf_data_251027.m: Main script to run the statistical analysis
• function_analyze_mnf_data_251027.m: Function called in the main script to run the measurements and statistics on all MNF maps and masks
• statistics_data.m: Sub-function used to calculate statistical values (mean, median, std, etc.)

results

This folder contains the sub-folders of results saved after running the data analysis using the main Matlab script run_analyze_mnf_data_251027.m.

Each sub-folder of results contains the following sub-sub-folders:

• mat: Final .mat file of the results when running the analysis using multiple N_ROI and N_REF values
• pdf: Final figures in .pdf format
• png: Final figures in .png format
• xls: Final tables of all measurements in .xlsx format

RUN THE DATA ANALYSIS

NOTE: Run the code from within the matlab folder. No need to add the folders in the MATLAB path.

Statistical data analysis using this code can be performed in two ways:

• Statistical analysis for a single value of N_REF and N_ROI. The example presented in the paper corresponds to N_REF = 80% and N_ROI = 90%.
• Statistical analysis for multiple values of N_REF and/or N_ROI. The example presented in the paper corresponds to N_REF = 5% to 100% by steps of 5% (20 values), and N_ROI = 5% to 100% by steps of 5% (20 values). In that case, three statistical methods are applied to estimate the final mean and standard deviation values of the MNF metrics: Standard, Jackknife, and Bootstrap.

Statistical analysis for a single value of N_REF and N_ROI

• Open run_analyze_mnf_data_251027.m, in section parameters:

  • Set param.ref_max_values_pc to a single value (in %), to select the N_REF maximum values from the ROI used to measure the mean PD and TSC references (in EYE or CSF, EYE is selected by default) for subsequent PD and TSC quantification in whole brain.
    Example: param.ref_max_values_pc = 80;.
  • Set param.roi_max_values_pc to a single value (in %), to select the N_ROI maximum values from the ROI used to measure mean and standard deviation in different ROIs (EYE, CSF, GM, WM).
    Example: param.roi_max_values_pc = 90;.
  • Other parameters can also be changed in function_analyze_mnf_data_251027.m, section analysis parameters, such as EYE or CSF reference for quantification of PD and TSC.

• Run run_analyze_mnf_data_251027 in the Command Window in Matlab. This will:

  • Run the analysis for a single value of N_REF and of N_ROI.
  • Calculate the mean and standard deviation of all MNF maps in all ROIs for each subject.
  • Calculate the mean and standard deviation over all subjects of all mean values calculated above.

• All results are saved in a sub-folder in the results folder, with name format mnf_analysis_refN_REF_roiN_ROI_date_time.

Statistical analysis for multiple values of N_REF and/or N_ROI

• Open run_analyze_mnf_data_251027.m, in section parameters:

  • Set param.ref_max_values_pc to a range of values (min:step:max, in %), to select the N_REF maximum values from the ROI used to measure the mean PD and TSC references (in EYE or CSF, EYE is selected by default) for subsequent PD and TSC quantification in whole brain.
    Example: param.ref_max_values_pc = 5:5:100; for selecting values from 5% to 100% by steps of 5% (20 values in total).
  • Set param.roi_max_values_pc to a range of values (min:step:max, in %), to select the N_ROI maximum values from the ROI used to measure mean and standard deviation in different ROIs (EYE, CSF, GM, WM).
    Example: param.roi_max_values_pc = 5:5:100;, for selecting values from 5% to 100% by steps of 5% (20 values in total).
  • Other parameters can also be changed in function_analyze_mnf_data_251027.m, section analysis parameters, such as EYE or CSF reference for quantification of PD and TSC.

• Run run_analyze_mnf_data_251027 in the Command Window in Matlab. This will:

  • Run the analysis for all combinations of the multiple values of N_REF and of N_ROI. This will result in:
    • Datasets of N_REF x N_ROI measurements of the mean values for PD and TSC for all ROIs in each subject.
    • Datasets of N_ROI measurements of the mean values for all ¹H and ²³Na relaxation times for all ROIs in each subject.
  • Calculate the mean and standard deviation of all MNF maps in all ROIs for each subject according to 3 statistical methods:
    • Standard: The overall mean and standard deviation is calculated over all mean values contained in each dataset of size N_ROI (relaxation times) or N_REF x N_ROI (PD and TSC) for each subject, using the standard averaging method (no resampling).
    • Jackknife: Jackknife resampling is applied to each dataset of mean values of size N_ROI (relaxation times) or N_REF x N_ROI (PD and TSC). The mean and standard deviation values are calculated for each resampled dataset, and an overall mean and standard deviation of the mean values of all datasets is finally calculated for each subject.
    • Bootstrap: Bootstrap resampling (or bootstrapping) with replacement, with N_btsp bootstraps (N_btsp = 1000 by default) is applied to each dataset of mean values of size N_ROI (relaxation times) or N_REF x N_ROI (PD and TSC). The mean and standard deviation values are calculated for each resampled dataset, and an overall mean and standard deviation of the mean values of all datasets is finally calculated for each subject.
  • Calculate the final mean and standard deviation values over all subjects of all overall mean and standard deviation values calculated with the 3 methods.

• All results are saved in a sub-folder in the results folder, with name format mnf_analysis_final_N_REF_xN_ROI_date_time.

REFERENCES

[1] G.G. Rodriguez, L. O'Donnell, Z. Yu, M. Cloos, G. Madelin. Multinuclear fingerprinting (MNF): high-resolution simultaneous proton/sodium MR fingerprinting. Proceedings of the Annual Meeting of the International Society for Magnetic Resonance in Medicine (ISMRM), Abstract #0579, Singapore, 2024.

[2] L. F. O'Donnell, G.G. Rodriguez, G. Lemberskiy, Z. Yu, O. Dergachyova, M. Cloos, G. Madelin. Correlation-weighted ²³Na magnetic resonance fingerprinting in the brain. NMR Biomed 38(11), e70150, 2025.

[3] Z. Yu, D. K. Sodickson , G. Madelin, M. A. Cloos. Simultaneous proton magnetic resonance fingerprinting and sodium MRI. Magn Reson Med 83, 2232-2242, 2020.

[4] Z. Yu, S. Hodono, O. Dergachyova, T. Hilbert, B. Wang, B. Zhang, R. Brown, D. K. Sodickson , G. Madelin, M. A. Cloos. Simultaneous 3D acquisition of ¹H MRF and ²³Na MRI. Magn Reson Med 87, 2299-2312, 2022.

[5] G. G. Rodriguez, Z. Yu, L. F. O'Donnell, L. Calderon, M. A. Cloos, G. Madelin. Repeatability of simultaneous 3D ¹H MRF/²³Na MRI MRI in brain at 7 T. Sci Rep 12, 14156, 2022

[6] G. G. Rodriguez, Z. Yu, S. Shaykevich, L. F. O'Donnell, L. Aguilera, M. A. Cloos, G. Madelin. Super-resolution of sodium images from simultaneous ¹H MRF/²³Na MRI acquisition. NMR Biomed 36, e4959, 2023.

FUNDING

National Institutes of Health (NIH) grant R01EB0264561.

NOTE ON DATA ANONYMIZATION

All shared data is completely anonymized.

Data was first saved and automatically de-indentified as a raw data (.dat) file from the scanner, images were then reconstructed and post-processed in MATLAB as number arrays (and where all MR sequence header and metadata information is lost), and finally saved as nifti (.nii) files in MATLAB.

Following the Data Management and Sharing (DMS) Plan from NIH, the shared data in this repository does not contain any Personal Health Information (PHI) such as subject identifiers, medical information, demographics nor biometrics. Moreover, during the post-processing step, the skin and skull features were removed such that facial and osseous features of the volunteers cannot be retrieved from the data.