mri_reface Releases

mri_reface executable 0.3.5-Mac

cgschwarz@www.nitrc.org (Christopher Schwarz) — Thu, 12 Dec 2024 15:03:00 GMT

0.3.5 ASL and PD support, Mac build, incremental improvements * Feature: Support for ASL (perfusion) images. Intended for Siemens PCASL and Philips PASL. Not needed for older PASL and GE PCASL sequences. * Feature: Support for PD-weighted images. This uses the same template and settings as T2-w images. Dual-echo T2/PD, PD/T2 are also supported. Thanks Cally Xiao for testing. * Enhancement: MEGRE MRI: Allow face intensity and noise matching to vary across volumes/echoes. * Feature: New "coregistered4D" option: Allows overriding default assumptions about which image types are coregistered and which are not, e.g. if multi-echo MRI are acquired over time, or PET frames were pre-aligned during preprocessing * Feature: New options "TIVTolerance" and "TIVToleranceOffset" to adjust the boundary between brain and face if necessary (e.g. if outputs changed a few voxels of brain, but were otherwise good, this may help). * Feature: New Mac build (non-Docker only; Matlab does not support building Docker images for non-Linux). This has the same requirements as current Linux builds: Matlab2022a runtime (free), reg_aladin (niftyreg), ANTS. Tested on MacOS Sequoia 15.1.1 with an M2 "Silicon" chip. DICOM support was tested with dcm2niix (20241208 from Macports) and Python (3.10.16 from Macports) with modules pydicom 3.0.1 and nibabel 5.3.2. * Fix: Use a new and smaller mask for sampling air, preventing errors in scans with extra-large heads and lots of motion. * Docker: Optimized the Docker image to 5.4G unzipped / 2.3G zipped. 0.3.3 was 7.5G / 5G respectively. * ADIR_nii2dicom: ignore .txt and rawdata (non-image/exam card) DICOM in DICOM folders * ADIR_nii2dicom: improved handling for DICOM where some tags are inconsistently present/absent. They will be considered inconsistent and will not be copied. * ADIR_nii2dicom: Add support for dropping frames from output DICOM, or ignoring DICOM from frames already dropped from the input nii * ADIR_nii2dicom: Handling for DICOM with StackID and DimensionIndexSequence/Values (were needed for ASL)
 0.3.4 (not publicly released) Additional PET image types support, improved PET noise models, and minor enhancements * Enhancement: Improved noise-matching models in the replacement face for PET input types. Old noise added to NF PET, especially in air, were too much. This now better matches input images. * Feature: Added support for "early" and full-dynamic PET (treated the same as each other). These activate automatically when given PET inputs with >=10 frames   * For these, the last frame is used as the reference, instead of the first, and frames at the start of the sequence with significantly less signal than those in the middle of the sequence are skipped from de-facing (copied as-is). These tweaks adjust for the presence of frames in the first few minutes that typically lack enough signal to warrant de-facing or to use as an anatomic refrence * Feature: Added support for PET tracers: PI2620 (treated like FTP) and NAV4694 (treated like PIB) * Feature: Added support for BIDS image type names (e.g. "T1w") in automatic image-type guessing * Fix: Prevent some crashes when input frames are empty (all zeros) * ADIR_nii2dicom: minor enhancements and fixes

mri_reface executable 0.3.5

cgschwarz@www.nitrc.org (Christopher Schwarz) — Thu, 12 Dec 2024 14:56:00 GMT

mri_reface Docker 0.3.5

cgschwarz@www.nitrc.org (Christopher Schwarz) — Thu, 12 Dec 2024 15:01:00 GMT

CHANGES 0.3.5 ASL and PD support, Mac build, incremental improvements * Feature: Support for ASL (perfusion) images. Intended for Siemens PCASL and Philips PASL. Not needed for older PASL and GE PCASL sequences. * Feature: Support for PD-weighted images. This uses the same template and settings as T2-w images. Dual-echo T2/PD, PD/T2 are also supported. Thanks Cally Xiao for testing. * Enhancement: MEGRE MRI: Allow face intensity and noise matching to vary across volumes/echoes. * Feature: New "coregistered4D" option: Allows overriding default assumptions about which image types are coregistered and which are not, e.g. if multi-echo MRI are acquired over time, or PET frames were pre-aligned during preprocessing * Feature: New options "TIVTolerance" and "TIVToleranceOffset" to adjust the boundary between brain and face if necessary (e.g. if outputs changed a few voxels of brain, but were otherwise good, this may help). * Feature: New Mac build (non-Docker only; Matlab does not support building Docker images for non-Linux). This has the same requirements as current Linux builds: Matlab2022a runtime (free), reg_aladin (niftyreg), ANTS. Tested on MacOS Sequoia 15.1.1 with an M2 "Silicon" chip. DICOM support was tested with dcm2niix (20241208 from Macports) and Python (3.10.16 from Macports) with modules pydicom 3.0.1 and nibabel 5.3.2. * Fix: Use a new and smaller mask for sampling air, preventing errors in scans with extra-large heads and lots of motion. * Docker: Optimized the Docker image to 5.4G unzipped / 2.3G zipped. 0.3.3 was 7.5G / 5G respectively. * ADIR_nii2dicom: ignore .txt and rawdata (non-image/exam card) DICOM in DICOM folders * ADIR_nii2dicom: improved handling for DICOM where some tags are inconsistently present/absent. They will be considered inconsistent and will not be copied. * ADIR_nii2dicom: Add support for dropping frames from output DICOM, or ignoring DICOM from frames already dropped from the input nii * ADIR_nii2dicom: Handling for DICOM with StackID and DimensionIndexSequence/Values (were needed for ASL)
 0.3.4 (not publicly released) Additional PET image types support, improved PET noise models, and minor enhancements * Enhancement: Improved noise-matching models in the replacement face for PET input types. Old noise added to NF PET, especially in air, were too much. This now better matches input images. * Feature: Added support for "early" and full-dynamic PET (treated the same as each other). These activate automatically when given PET inputs with >=10 frames   * For these, the last frame is used as the reference, instead of the first, and frames at the start of the sequence with significantly less signal than those in the middle of the sequence are skipped from de-facing (copied as-is). These tweaks adjust for the presence of frames in the first few minutes that typically lack enough signal to warrant de-facing or to use as an anatomic refrence * Feature: Added support for PET tracers: PI2620 (treated like FTP) and NAV4694 (treated like PIB) * Feature: Added support for BIDS image type names (e.g. "T1w") in automatic image-type guessing * Fix: Prevent some crashes when input frames are empty (all zeros) * ADIR_nii2dicom: minor enhancements and fixes

mri_reface executable 0.3.3

cgschwarz@www.nitrc.org (Christopher Schwarz) — Fri, 17 Nov 2023 21:25:00 GMT

0.3.3 Minor enhancements and bugfixes * mri_reface now detects when an input .nii has inconsistent geometry (Sform vs Qform mismatch) and prevents this from causing invalid results by attempting to choose the best one based on the transform intent codes. Regardless, the nii geometry of the input image is retained in the de-faced output, even if it was inconsistent (we fix it internally, then put it back to however it was). * If given a 4D nii of a PET type, automatically split them into separate frames and de-face them together to allow for motion between frames. Previously, the best practice was to split and merge them externally. * Updates to ADIR_nii2dicom   * Fix a crash when filepaths contained certain sequences   * Correct behavior of the NumberOfSlices tag to reflect the number of slices in each volume, not the total across volumes   * Increased compatibility for concatenated DICOM (usually Siemens Vida)

mri_reface Docker 0.3.3

cgschwarz@www.nitrc.org (Christopher Schwarz) — Fri, 17 Nov 2023 21:45:00 GMT

mri_reface executable 0.3.2

cgschwarz@www.nitrc.org (Christopher Schwarz) — Wed, 26 Apr 2023 19:21:00 GMT

0.3.2
Matlab updates, improved DICOM support, minor enhancements, and first Docker release
* Matlab runtime version update: now uses and requires matlab 2022a
* First release of a Docker image (separate download/file)
* run_mri_reface.sh now supports a workflow where you can input a directory containing a single DICOM series, instead of a .nii image. This will be converted using dcm2niix, de-faced, converted back to dicom, and marked de-identified. This requires dcm2niix to be installed, and a python environment with pydicom and nibabel.
* Add a -faceMask option to specify an alternative set of regions to replace. This image MUST be in the voxel space of MCALT_FaceTemplate_T1.nii. Voxel value 1 = face; 2 = air behind the head potentially containing wraped face-parts; 3 = ears. Warning: Using this option may produce de-faced images that do NOT offer adequate protection from re-identification.
* For images with larger voxels (e.g. PET and 2D MRI), slightly increase the boundary around the TIV (i.e. brain) that will not be replaced
* Save the _Affine.txt earlier during processing, for de-bugging and resuming purposes.
* We now recommend niftyreg version 1.5.x or better, due to user reports of worse performance with 1.3.x.
* Minor updates to ADIR_nii2dicom

0.3.1 (internal release)
Minor improvements
* Improved intensity normalization mask for 2D FLAIR (type AFL or 2FL)
* Random noise in replacement face now uses a fixed seed for reproducibility
* If an output Affine.txt is found in the output directory, use it (resume processing from this step). Previously, we would only use affine+warp sets found together but would ignore a lone affine.

mri_reface Docker 0.3.2

cgschwarz@www.nitrc.org (Christopher Schwarz) — Wed, 26 Apr 2023 20:47:00 GMT

mri_reface Docker:
This Docker image replaces face and ear imagery in brain MRI/PET/CT with an average face, to help prevent potential re-identification

By: Christopher G. Schwarz and Carl M. Prakaashana, schwarz.christopher@mayo.edu
For details, please see this paper and cite it if you use it in your work:
Christopher G. Schwarz, Walter K. Kremers, Heather J. Wiste, Jeffrey L. Gunter, Prashanthi Vemuri, Anthony J. Spychalla, Kejal Kantarci, Aaron P. Schultz, Reisa A. Sperling, David S. Knopman, Ronald C. Petersen, Clifford R. Jack. Changing the face of neuroimaging research: Comparing a new MRI de-facing technique with popular alternatives. In: NeuroImage 231, 2021. https://doi.org/10.1016/j.neuroimage.2021.117845.
For PET/CT support: Christopher G. Schwarz, Walter K. Kremers, Val J. Lowe, Marios Savvides, Jeffrey L. Gunter, Matthew L. Senjem, Prashanthi Vemuri, Kejal Kantarci, David S. Knopman, Ronald C. Petersen, Clifford R. Jack. Face recognition from research brain PET: An unexpected PET problem. In: NeuroImage 258, 2022. https://doi.org/10.1016/j.neuroimage.2022.119357
This software works only with Docker environment installed. First, run ''''docker load < mri_reface_docker_image''''. You should then run it through the included run_mri_reface_docker.sh rather than entering the Docker environment directly. The Docker image includes all dependencies of mri_reface and does not require any software be added to the user''''s system other than Docker.
mri_reface Docker allows Nifti (.nii) or DICOM inputs. If the user provides a .nii file, mri_reface Docker will output a refaced .nii file and before/after face renders. If the user provides a DICOM directory as an input, in mri_reface Docker will convert the image into .nii, reface the .nii image, convert the .nii image back to DICOM (copying all unaffected tags from the original DICOM to the new DICOM) and tag the DICOM files as being de-faced. The output for DICOM input include the refaced .nii, before/after face renders, and the refaced DICOM files. The direct-DICOM workflow is NOT recommended for PET inputs, as it will not account for motion across frames.
For .nii inputs, the input path should be to the .nii file itself. For DICOM inputs, the input path should be to a directory, this directory should contain ONLY the DICOM files for the image (a single series).
This work was supported by: NIH grants R01 AG068206, U01 AG006786, P50 AG016574, R01 AG034676, R37 AG011378, R01 AG041851, R01 NS097495, R01 AG056366, U01 NS100620; The GHR Foundation; The Elsie and Marvin Dekelboum Family Foundation; The Alexander Family Alzheimer''''s Disease Research Professorship of the Mayo Clinic; The Liston Award; The Schuler Foundation; and The Mayo Foundation for Medical Education and Research. We also gratefully acknowledge the support of NVIDIA Corporation with the donation of a Quadro P6000 GPU used in generating 3D facial reconstructions for this research. We also thank AVID Radiopharmaceuticals, Inc., for their support in supplying Flortaucipir precursor, chemistry production advice, and FDA regulatory cross-filing permission and documentation needed for this work.
See LICENSE.txt for license restrictions
Copyright 2020-2023 Mayo Foundation for Medical Education and Research

mri_reface executable 0.3

cgschwarz@www.nitrc.org (Christopher Schwarz) — Fri, 08 Jul 2022 20:42:00 GMT

0.3
Add PET/CT support, multi-frame and multi-image support, and noise matching
* New templates for PIB, FBP, FTP, FDG, CT
* Support for multi-frame (4D .nii) inputs, and multiple (separate .nii) inputs
* Attempt to match noise intensity and PSF of input image, and apply this to the template image (on by default, can be disabled with -matchNoise 0). This visually improves images and was necessary for support of unfiltered dynamic PET images; it has measurable but very small benefits to downstream volume measurements from T1 MRI as well.
* Coincides with the first release of our accompanying ADIR_nii2dicom utility
* Add -altReg parameter to try an alternate registration scheme when the original fails
* Add -threads parameter to control multithreading during reg_aladin and antsApplyTransforms, but it has minimal effect. Default is 1 (they used to use the maximum available)
* Fixes for users who rely on non-standard LD_LIBRARY_PATH for dependencies to function (MUST use run_mri_reface.sh)
* Output deFaced.nii will now match the bit depth and scl_slope/scl_inter of the original input .nii. Previously they would be upsampled to floats.
* Primary outputs of this version would be identical to 0.2 for MRI inputs if -matchNoise 0

mri_reface executable 0.1.3

cgschwarz@www.nitrc.org (Christopher Schwarz) — Mon, 21 Jun 2021 17:17:00 GMT

Results from this version match our 2021 NeuroImage publication "Changing the Face of Neuroimaging...". This version was also used in the public release of baseline MRI from the A4 study. Choose this version if you want to reproduce those specifically. Otherwise, if you just want to de-face data, choose the latest version instead.

mri_reface executable 0.2

cgschwarz@www.nitrc.org (Christopher Schwarz) — Mon, 21 Jun 2021 17:24:00 GMT

Prevention of face recognition is substantially improved from 30% to 7%. Effects on downstream GM measures are slightly larger than with 0.1 but remain far below scan-rescan effects and competitive with other leading de-facing algorithms.