FireVoxelPro Transition Announcement

After many years of free academic distribution, the FireVoxel software distribution model is changing.

Future maintained releases will transition to FireVoxelPro, an independently maintained licensed edition of FireVoxel developed, licensed, and distributed by Artem Mikheev.

The goal of this transition is to keep the familiar FireVoxel workflow available while making continued maintenance, compatibility updates, bug fixes, support, and future development sustainable.

Current FireVoxel release timeline

The current FireVoxel release will begin displaying expiration notices on June 1, 2026, and will expire on July 1, 2026.

Beginning July 1, 2026, future maintained releases will be distributed as FireVoxelPro under a new licensing/support model.

Why this change is happening

FireVoxel has been developed and maintained continuously over many years. For a long time, users were able to benefit from free releases and regular updates. However, ongoing maintenance of a specialized scientific imaging platform requires substantial engineering time, testing, compatibility work, documentation, and user support.

The previous free academic distribution model is no longer sustainable. FireVoxelPro is intended to determine whether there is enough user, lab, and institutional support to keep the software actively maintained as an independent project.

What is FireVoxelPro?

FireVoxelPro is the independently maintained licensed edition of FireVoxel.

The user interface and general workflow will remain familiar to existing FireVoxel users. The goal is continuity, not disruption. FireVoxelPro will include maintained releases, performance improvements, bug fixes, and new general capabilities over time.

FireVoxelPro is not affiliated with, endorsed by, or supported by NYU or NYU Langone Health.

Licensing model

The expected introductory research licensing model is:

Individual research license:
$2,000/year for one active user/workstation

Small-lab license:
$5,000/year for up to 5 active users and up to 5 activated workstations

Larger lab, center, or institutional licenses:
Quoted separately depending on number of users, workstations, and support needs

 1. Duke breast database: implemented the label map merge for Full breast and FGT Vessel

2. Implemented new median sort function and updated entire database.

3. Implemented rTSNR moco evaluator. 

4. Implemented the procedure to extract AdiposeZone and FgtHalo from Duke's label maps. 

5. Substantial speed up in Similarity calculation (XCORR) due to parallel re-arrangement.

6. FLEXR Registration\MoCo now run on Memory-mapped file requiring a fraction of RAM from previous versions.

7. Fixed progress defect in FLEXR MoCo 

8 Sped up the ViterbiMatch function in FLEXR.

9. Corrected defect in COnvolution-based  NW  sparse-to-dense functionality. 

10. FLEXR : added option to ForceInslice (displacement), so it is a 2D flow field. 

Build 483 is released. 

 1. Initial integration of FastSurfer functionality under GUI. Display of 100+ ROIs after segmentation.

2. x6 speed up of zooming of ROIs measured over the FilmView

3. x10 speed up of Quasi-integer check and RealToInteger conversion both of each were a part of the DICOM load. 

4. Major x10s speed up of FLEXR  deformable registration (sparse-to-dense flow field module).

 Build 482 is released

1. Zlib is integrated for   nii.gz,  mgz, and nrrd support. Correspondingly 7zip.exe and DLL are removed.

2. NRRD file format is supported through File>Open Image  (where nifti etc). This is supported for the integer file reading. NRRD is very often used for ground truth storage (f.e. Duke database). Read  only.

Build 481 is released. 

 1. Fixed Volume>Set Orientation GUI function.

2. Fixed orientation loading in MGH files 

Build 479 is released. 

 1. Correction "V[t]-V[1] subtraction function.

2. ROI layers to atlas map function : enabled 4D operands. 

3. Fixed defect: when new ROI is created in Layer Control Dialog (Single slice), the Z-resolution now matches the base layer (isotropic before fix). 

4. Improved parameters in the Lung MRI workflow Segmentation by Registration.

5. Improved BM1 workflow so it works with .gz files and optionally without ROIs.

Build 478 is released. 

 1. Implementing the partial labeling policy for breast tissues(9) segmentations from GRASP.

2. Implementing the custom AdamW initializer.

3. Moved HardDICE calculation logic to CPU-non-Tensor for more detailed control. 

4. Implemented GUI function for V[t]-V[1] subtraction (BPE) calculation. 

Build 477 is released 

 1. Finalized FiLM categories in the Unet. Implemented "Time as Category" approach and validated on H130 set.

2. Reimplemented inference for the above, resulting segmentation compensates the motion. 

3. Restored the GRASP-4D training functionality with the same BrUnet2D network used for H130.

 4. Fixed defect with 3-projection view

Build 476 is released. 

 1. Fixed defect in Surface point detection for wave propagation. This was also rippling into BiCal processing.

2. Implemented corner inpainting during the affine augmentation based on wavefront averaging.

3. Implemented the Augmentation mask which changes during augmentation and is used for restricting the cost calculation. 

4. implemented FiLM mechanism: categories incorporated into U-net learning. Implemented "Times as Category".

5. Implemented the new model and integrated under  "BRUNET_H130" model menu

Build 475 is released. 

1. Adapted MirrorX and MirrorY functions for multi-channel  augmentation.

2. Fixed a substantial defect in VA_IsInSliceProj.  It was causing a crash under certain scroll parameters.

3. Fixed defect in GetResolution function.

4. Breast Unet: implemented Tversky cost.

5. Corrected 3D Separable convolution defect affecting the N3.

 

Build 474 is released.

 1. Corrected a defect in DlgVolInfo: the Rows\Columns\Bpp fields were incorrectly displayed.

2. Changed the C++ API for memory handling functions to template instead of legacy defines.

3. Simplified requirements for the .TIME file for loading 4D (Nifti,Midas, etc). Now only a few fields are required.

4. Fixed defect RoiStats4D when no ROI was present.

Build 473 is released.

 1. Fixed the defect: when a single invalid multiframe file is present in the subfolder - the whole DICOM load failed without diagnostics. Now problematic file names are shown to the User who has an option to proceed without them.

2. Fixed defect in sorting calling contract in DICOM Tree record sorting (RawTime and RawTimeSerNumber).

3. Fixed several memory leaks in DicomMakeTree.cpp

4. DicomMakeTree.cpp : C++17 compliance: removed binary operators in parallel sorting.

Build 472 is released

 1. Allowed a reasonable display of DICOMs when the SliceSpacing and Slice thickness  = 1e60 (VOIDVOX)

2. Fixed bug in FVXDCM_GetPixelDataInfo function.

3. Fixed bug in FVXDCM_MinimizeDicomHeader function.

4. Fixed bug in FVXDCM_SaveSliceDirect function.

5. Fixed microseconds (UINT16) truncation bug.

6. Fixed DCM_PixelSpacing defect with switched Vx,Vy order.

7. FVXDCM  DJ codecs clean-up is fixed.

Build 471 is released.

1. Implemented FlexDim design for the Unet 2.5D segmentation network.

2. Dicom dialog box: implemented 2 options for working with "Combined Studies", so they can be copied

3. Fixed logic errors in "Enumerate Tree Items with Checked Ancestors". This was causing the problem of duplicated files when copying the DICOM subtree.

4. Removed the "Multiframe warning" from DICOM Tree to Folders function.

FireVoxel Build 470 is released

 1. Fixed long-standing error when after multilayer volume rotation often the message would come up between layer resolution mismatch.

2. Expanded and debugged the "FVX folder to  Major orientation batch" GUI function.

3. Started the DUKE100H preparation workflow under a separate submenu,

4. FLEXR3D\MC:  modified procedure how keypoints are constructed. This was to fix a situation of very thin images coming out with 0 keypoints.

5. Updated ABT file set

6. Batch function for FVX folder to Unify Times and Single Slice.

7. Fixed defect when in newer DICOMs the PixelSpacing information is located within nested Sequences. Same for SliceThickness.

8. Various corrections for IVIM model #17, related to the model-unsuitable voxels.

Build 469B is released.

 1. WML processing: scaled INTEGER signal to Max=10,000. This improves the operations and reduces the effects of interpolations.

2. Fix  VolMul function.

3. Code work in VolumeSupport.cpp

Build 469A is released.

 1. FVX expiration date was changed.

2. WML batch processing:  default algorithm changed to #2 everywhere.

3. First WML batch function was renamed to "NIFTI folder: WML batch processing".

3. Fixed the "DICOM folder: WML batch processing" so the function accpets  {0,1,2} case subfolders per case (2 subfolders correspond to FLAIR+MPR)

3. Fixed the "FVX folder: WML batch processing" (failed due to a new QuickSave function)

Build 469 was released.

 1. Implemented a batch function to take a folder of FVX files. Each containing a 4D volume and an ROI.  Batch motion correction is done and all results are written to the Target folder.

2. Fixed 2D Raster Skeleton algorithm.

3. Fixed the rotation conventions in the XYZ rotation and the Orthogonal rotations.

4. Batch function to convert a folder of FVX files into a specified Major orientation.

5. Ongoing work for the Unet improvements: variable dimensions, categories, etc.

Build 468 is released.

 1. Breast Unet 2.5D:  adapted current augmentatiton to support the channel dimension

2. Implemented function to split a folder of NIFTI files into Left\Right images (support Breast Unet).

3. Release of AI integrated model for GRASP4D breast segmentation

4.  Release of AI integrated model for VIBE4D breast segmentation

 Build 467 is released.

1. Re-engineering of Model0 and Model1, calculations and dialog.

a) Changed Model#1   Start\Length hyperparameters.

b) Model#1 name changed to "AUC"

c) MaxVal output parameter moved to Model0.

d) Removed warning about change in model hyperparameters (due to software update).

2. ROI>Threshold by Signal interval:  Now works on 4D volumes by returning the

mask for the entire 4D volume.

3. Initial implementation of the Breast 4D volume segmentation based on U-net.