Implementing the "Sokoloff Model with the perfusion-derived Input function"

Verified quality of registration of GRASP-4D to PET4D.  First timepoint of GRASP is cropped and the regular   Applications>PET>Register to Anatomy function is used.  Result is below in several projections.  PET (in RED) is overlayed on top of  GRASP (blue).

Notice substantial (~10 degree) rotation angle between  GRASP and PET.

When saving TAC to file or Clipboard, and appropriate Header is placed before the 2-column data. In the following format    "Dynamic Component (Dynamic Units)  - TAB - value"  so it is suitable for an easy incorporation to Excel.

Enabled saving the "Fitted Curve" from the RoiStats4D dialog.  When user chooses "Save (to file)" or "Copy (to clipboard),  Dialog choice appears that allows to select either "Active ROI" or "Fitted" curve.

FireVoxel Build 128F is released.

1. Implemented initial variant of PET 4D correcction. It is located under MainMenu>Applications>PET>Register 4D (Eliminate motion)
2. In ViewTAC, and RoiStat4D dialog, when the curve is drawn the current Dynamic Index node is marked with the alternative color.
3. Correct  "Edge Constrained Smoothing" function so the "Fill Void Voxels"  option works correctly. Removed redundant code.
4. Corrected the defect in the Progress indicator "with User interruption" function.  This was causing problems that manifested as DICOM loading problems.
5. Corrected the PET-DICOM loading failure that appeared in Build 128D.
6. Fixed the error with PET frame timing due to a simple typo error.
 

Corrected the defect in the Progress indicator "with User interruption" function.  This was causing problems that manifested as DICOM loading problems.

Fixed the error with PET frame timing due to a simple typo error.

Corrected the PET-DICOM loading failure that appeared in Build 128D.

Conducting a short evaluation for the next generation of BiCal non-uniformity conversion tool.  Focus is on using the Gradients instead of Edges for delineating regions of homogeneous tissue.

Correct  "Edge Constrained Smoothing" function so the "Fill Void Voxels"  option works correctly. Removed redundant code.

Prototyping a new IDIF function variant, that would explicitly incorporate an explicit anatomy constrained:  namely a tube with the width of the vessel.  User specifies one of the {X,Y,Z} prevalent
directions.

In ViewTAC, and RoiStat4D dialog, when the curve is drawn the current Dynamic Index node is marked with the alternative color.

Implementing and initial variant for PET  4D motion compensation.  This is based on Dr. DeLeon comment during IDIF meeting.  Additionally a substantial rotation is observed (~2 degrees) betweeen MRI and some of the PET timepoints.
Standard 4D registration is used with some improvements to determine PET forground mask. Main challenge here is the speed of the registration due to the high PET timepoint count.

Investigating the reslicing problem after the registration.

Correcting the IDIF function,  for Axial-SingleSlice variant.  It should behave correctly in {NoSeed,SingleSeed,MultiSeed} modes.

FireVoxel 128D is released with the   4DPET-MRI scenario fixes.

4DPET-MRI registration:   Corrected the workflow so user could choose the "Best" or maximum resolution evaluated individually at each axis.  Registration result is good.  Importantly, notice a non-trivial rotation angle between MRI and PET.  On this screenshot they are registered, but observe the MRI slanted margins, indicating ~2 degrees relative rotation as determined by the registration.

Helping to process the MRI+PET case obtained from Dr. Wahle.   MRI is in Axial, which is different  from the documentation.

Updating the IDIF function for a case when no MACRO-seed is present at all and instead user chooses simply the slice on the Axial projection.

Play 4D function:

• Rewind if we are on the last frame
• Use the default the interval to the last one used
• Indicate the use of real intervals

Investigating a substantial memory residual remaining after the PET-to-MRI registration  ABT subtest.

Updating the Automatic Build Test   for 4D-PET and MRI co-registration.

Dialogs   {AutoFocus3D,AutoFocus4D} :  removed "URAL Interpolation" parameter with the purpose of simplifying registration User Interface.   URAL Interpolation = true, in all registrations.

Documenting the changes in the PET workflow after the change to the Axial projection.  Updating the manual.

Found solution for the registration of the Femur\bone ROIs. This is for the same patient, longitudal study.

Updating and restructuring the registration workflow to principal Axial-projection and accommodate new requirement to either reconstruct to Anatomy or PET resolution.

FireVoxel Build 128B is released.

Made Parametric Calculation call smarter - in case there is a single 4D layer is present, it is selected automatically and User does not have to make it Active.

Major code refresh for:
Dynamic Processing dialog box
Dynamic parametric map calculation - UI Application level
All models in Dynamic Processing DLL are revisited for
parameter consistency.

Parametric modelling dialog layout is rearranged to accommodate longer labels automatically generated for Input Functions.

RoiStats4D dialog:  when TAC for the ROI is obtained, its label is automatically formed consisting of the Roi Name + ROI size in mm3.  This allows user to avoid manual typing in when importing the Input Function inside the Parametric Modelling dialog.

Dialog Roi Stats4D:  Added the "Show Median"  checkbox (as apposed to "Show Average") .  Roi TAC is shown either in Median or Average mode. Same choice for Saving and copying it to the Clipboard.

FireVoxel Build 128A is released.

RoiStats4D dialog:  when "Show All" checkbox is ON,  only Enabled layers will display the TACs.

Improved behavior in RoiStats4D function.  When F4 button is pressed and there are both 3D and 4D layers are present, 4D layer will always be chosen, even if 3D is Active.  This eliminates annoying messages "4D layer is required".

UI change:
Function key F3 is reassigned to be   a "Roi Stats 3D" function
Function key F4 remains a "Roi Stats 4D" function = SHOW TAC
Former key F3 is now F2 for the "Copy ROI info to clipboard" function.

The reason for the change is that often a dynamic image is overlayed
on top of static 3D, so getting TACs often pops up the "4D layer required"
which requires user to bring up the Layer Control box, change active layer
and so on.

Completely re-engineered the 1st Model parameters to make their meaning more clear.

Model Name   = "Signal measurements within time interval";
{ "Signal Spread","Mean Signal","Max signal","Signal StdDev","Sum of absolute signal changes",
"Turbulence","Max Turbulence","Turbulence Count" }

Integrated the "Process Interruption Feature" across FireVoxel.  For any process that displays the progress bar, if user press ESC button, function gracefully exits.  94 locations were modified, only few were tested but expected all to work correctly.

Substantial redesign of the Play4D function:  Movie plays in all available projections simultaneously,
dynamic coordinates, such as b-value, are displayed and movie sequence is in this active dynamic dimension order.

Additional feature,  movie is played in all available views of the selected document
(f.e. all orthogonal projections)

Integrating general UI functionality so that any process that outputs the progress bar may be interrupted by User pressing ESC button....

Fixing a problem with  DlgSpecifyDouble dialog box...

FireVoxel build 127B is released.

Play 4D movie function:  completely re-engineered, removed the defect.  Added a progress bar to the play function.  Additionally user can specify the interframe delay in seconds (real).

Get IDIF function User Interface:  provide a more intuitive user interface especially in the presence of multiple ROI layers, so that no "Ambiguous layer configuration..." message does not show up

After integrating time delay parameter, a new fitting result.

Sokoloff model:   Integrating the "Time Delay" fitting parameter. At the very least, this time delay parameter should be in [0,FrameTimeDelta] (=[0,15] sec) as this is the temporal resolution of TAC.

Updating the Sokoloff model.  During the demo to Fernando, observed that in "naive convolution" calculation upsampling time to 1 sec makes a big positive precision difference.  But this is computationally expensive and additional Upsampling Interval parameter will add to parameter list.
So instead attempting to upsample only the IRF ( Impulse Response Function) and integrate fast convolution technique from:

Fast and accurate voxel-by-voxel perfusion imaging using convolution models.  Artem Mikheev, Henry Rusinek.  Image Analysis - Parametric Mapping Session at the ISMRM 2008 Totonto, CANADA.

Revisiting Volume Smoothing function and ensuring it works correctly on all of combinations of:
{Real,Integer}x{3D,4D}x{ROI,no ROI}...

Fixed the problem with MainMenu>ROI>Threshold To ROI function.  Corresponding Build 127A is released.

Implementing the Sokoloff model with simple fitting by gridding+Amoeba....

FireVoxel Build 127 is released.

• Developed  the PET Workflow Quick Start manual.
• Produced the initial result for the Image Derived Input Function. 
• Implemented Vt (Logan) plot parametric model based on the newly available technology for the Image Derived Input Function from Carotid Artery.
• Corrected the defects that came up during the presentation to Fernando.  These two bugs are intermittent and are hard to reproduce.
• Implementing a new version of the "Blood Vessel Finder" dialog based on the initial discussions.
• Finalized 2nd generation of the IDIF (Image Derived Input Function) dialog. Principal addition is the introduction of L2-based measure for checking TAC similarity.  However, L2-smoothing was introduced in a way so blurring of the TAC is avoided.
• Extending the IDIF functionality so several seeds can be specified covering branches of the Carotid Artery to create larger ROI and thus more reliable Input Function. Since the IDIF function is very fast this is done simply by calling procedure several time and OR-ing corresponding ROIs to create the final one.
• Documented the latest version of the semi-automatic IDIF tool.
• Fixed the "Missing time information in save .IM file"  reported by Dr. Rusinek.  This was due to the time written in "absolute value" and using the scientific notation some important lower digits were truncated.  Switched to the regular "double float" format by using %lf option in fprintf().
• Fixed parametric map-related defect:  when  RoiStat3D  dialog display statistics, the VOID voxels are included into the calculation.  They should be excluded.
• Simplified the "Real Volume to Integer conversion" menu item to perform conventional rounding with the truncation to the specified bit length.'

Simplified the "Real Volume to Integer conversion" menu item to perform conventional rounding with the truncation to the specified bit length.

Fixed parametric map-related defect:  when  RoiStat3D  dialog display statistics, the VOID voxels are included into the calculation.  They should be excluded.

Fixed the "Missing time information in save .IM file"  reported by Dr. Rusinek.  This was due to the time written in "absolute value" and using the scientific notation some important lower digits were truncated.  Switched to the regular "double float" format by using %lf option in fprintf().

Documented the latest version of the semi-automatic IDIF tool.

Extending the IDIF functionality so several seeds can be specified covering branches of the Carotid Artery to create larger ROI and thus more reliable Input Function. Since the IDIF function is very fast this is done simply by calling procedure several time and OR-ing corresponding ROIs to create the final one.

Verified that the Macro Seed for the IDIF may be specified on any of the orthogonal projections.`It makes it possible to specify the Macro Seed on the Axial projection where Carotid artery can be found easily.

Finalized 2nd generation of the IDIF (Image Derived Input Function) dialog. Principal addition is the introduction of L2-based measure for checking TAC similarity.  However, L2-smoothing was introduced in a way so blurring of the TAC is avoided.

Implementing a new version of the "Blood Vessel Finder" dialog based on the initial discussions.

Fixed the registration mishap during the presentation.  After PET->MR registration, the time automatically advances to the first timepoint where PET signal is non-zero.

Correcting the defects that came up during the presentation to Fernando.  These two bugs are intermittent and are hard to reproduce.

Implemented Vt (Logan) plot parametric model based on the newly available technology for the Image Derived Input Function from Carotid Artery.

Implementing Vt (Logan Plot) suitable for FDG based on the latest IDIF availability.  Model Interface is almost identical to the PMOD.

Presently implementing support functions for manipulation Input Functions within the Dynamic Processing module.

Produced the initial result for the Image Derived Input Function.  This is developed for the PET FDG priority at the moment.  However, function is just a very general "Input Function Vessel Detector", so can be used in the DCE studies too.

Briefly, after PET is registered to MRI, workflow is as follows:

• User specified a very approximate rectangular region containing carotid artery.
• User specify a single "Vessel Homogenity" parameter as a percentage (f.e. 70%)
• Automatic process follows, the IDIF   TAC dialog pops-up for the inspection
• Dynamic Framework Input Function is populated automatically so the user can proceed immediately to kinetic models.

Additionally, FireVoxel provides a simple and fast tool helping User to find an approximate location of Carotid artery.

Have an algorithm outline for the PET FDG    Image derived Input Function.  Started UI implementation...

Implementing automatic OR possibly semi-automatic procedure for Image Derived Input Function (IDIF) extraction from FDG PET dynamic volume.

Continue work on the PET Workflow Quick Start manual....

Investigating the core functionality of DCMTK library for converting their internal DICOM header structure to the library-independent tree structure of the FireVoxel.

FireVoxel Build 126A is released.

Corrected the ADC  diffusion parametric model so that the fitting can be done even when only 2 timepoints are present.

Creating the "PET-workflow QuickStart manual"....

Official Build 126 is released to testing.

2.  Removed the "by-default" calculation of Source and Target noise in AutoFocus 3D\4D.

3.  Changed the linkage of the Dynamic Frame work DLL, so all the supporting DLLs can be used directly without redundant duplication.

4. Moved all Amoeba code into the PWR dll so it can be accessed by the Dynamic DLL.

5. Removed the IVIM with background model.

6. When dialogs RoiStat3D and RoiStat4D are acivated, the "ROI is empty" error message is displayed,  show the name of the ROI layer.

7.  Register sing Landmarks dialog:  fixed several problems in specifying the name of the Transformation  .vtf file.

8. "View TAC" dialog box:  added the "Voxel Radius" parameter so the TAC could be averaged over the cubic vicinity of the cursor.  This results in seeing smoother TACs. Parallelized to work smoothly over the volumes with the hight # of timepoints.

9.  "Landmark registration with autofocus" is changed to "Landmark registration + Voxel similarity", execution crash was fixed.

10. Fixed an intermittent error in the core Registration procedure, when registration image padding was calculated based on the URAL radius, while that radius was undefined due to the different registration measure used (i.e. non-URAL, f.e. MI etc)

11. Defect: crash in T2\T2* mapping model for the 4D abdominal volume provided by Henry. Fixed: the problem was in incorrect handling of timing data, when the Dynamic Component is different from TIME (ECHO in this case).

12. Fixed defect:   4D registration would not initialize for 4D Real-valued volumes such as PET.  Presently, 4D registration with Mutual Info was tested on 4D PET volume.

13. Fixed the  3D Edge Detector function to work properly on the Real-value volumes such as PET.

14. Fixed the  3D Texture Gradient function to work properly on the Real-value volumes such as PET.

15.  Problem with Alpha-combining the White-color ROI is due to that the White color 24-bit value of 0xffffff is also used as a transparency voxel code..

16. Added a warning message to 3D and 4D registration dialog boxes.  When the user have chosen the URAL-UNI registration measure, but the Sorce\Target Noise were not specified, the warning is displayed when user presses OK button.

17. Implemented averaging function for Signal Intensity and Real-valued volumes.User specifies the Time Index interval for averaging. Resulting REAL valued volume is overlaid on top of the original 4D.

18. Implemented the Mead-Nedler (simplex\Amoeba) optimization algorithm MDS (multi-direction search) variant.  This is inherently parallel algorithm, presently verified working in the serial workflow.

Presently: inserting parallel instructions to fully utilize MDS parallelism.

19. Enabling registration of REAL volumes using Mutual Information measure.  Presently the code is only working for 16-bit volumes (that was used to show PET->MPRAGE prototype).  Solution is to convert to 16-bit correctly while binning the voxel values for the MI.  This is also strongly related to a proper binning for REAL volumes, so it is addressed at the same time along with the corresponding binning interface  in the registration dialogs.

20.  In Registration Finetune, both 3D and 4D, adding a "Multipass" option corresponding to initiating Amoeba algorithm several times.  This optionally achieves higher registration scores at the expense of the execution speed. 

21. Creating a fully integrated workflow of 4D PET registration to MRI, so that user does not have to manually crop and change orientation of PET prior to registration.

22. Fixed defect:  when any type volume is displayed, if user presses ViewFilter for this volume, there is a fairly long wait and the "Compute Histogram" message is displayed.  This was fixed by removing Histogram recalculation and instead using compressed block header access without decompressing the blocks themselvves to find maximal voxel value of the volume.

23.  Corrected the logic of automatically finding an optimal window\center for displaying the image for all kinds of volumes (i.e. Real and Integer). Problem is in 4D cases, only the 1st timepoint is evaluated to reduce the processing time. But first or any other timepoint might contain almost no signals thus leading to a non-sense settings and strangely displayed volume.

24.  Fast loading of the native *.FVX file:  98% of time is spent    DCMTK::FindTagName function.  Result:   "Native" loading of the original DICOM 5.6GB dataset is now 0.7 sec.

25. Highly parallelized Volume cropping function (UI-level), so it runs much faster now.

26. Achieved final registration of PET to MPRAGE.   It is part of the ntegrated registration workflow that includes automatic cropping and orientation correction (from coronal to sagittal).

27. FilmView display:  for large images it is quite slow, with 4.8 sec to display all slices of 344x344x170 volume.  Fully parallelized with the x3.5 speed-up (down to 1.37 sec)

28. Eliminated the need for the user to specify the "anchor timepoint" during the PET to Anatomy registration.  Timepoint of Maximum average singnal is selected and the registration is performed to that timepoint to determine the required transformation for all PET timepoints.

29.  Implementing MRTM0, MRTM, MRTM2 reference model within the FireVoxel Dynamic Processing framework.

30. Fixed  the DICOM-ROI   save and edit problem reported by Alexandra Hotca.

31. Fixed several problems when trying to load a single DICOM slice directly from multi-slice DICOM tree.

32. Fixed defect: When ROI is saved to DICOM it is incorrectly displayed in Oped DICOM dialog Preview panel.

33. Fixed an error writing incorrect datatype in ANALYZE 7.5 format (for 16-bit volume)

34. Added the progress indicator to the FireVoxel document load. This is very helpful when loading very large volumes (such as PET)

35. When dialogs RoiStat3D and RoiStat4D are acivated, the "ROI is empty" error message is displayed,  show the name of the ROI layer.

36. In Dialog ROI  Stats 4D:  added a function to copy TAC to a clipboard in format so it could be easily pasted to Excel without saving to a file.  Also, changed the formatting of  TAC file to include \t to import easily to Excel.

Fully cleared and corrected the test logic for ABT on FireVoxel Build 126.

Added a warning message to 3D and 4D registration dialog boxes.  When the user have chosen the URAL-UNI registration measure, but the Sorce\Target Noise were not specified, the warning is displayed when user presses OK button.

Fixed the Phillips diffusion loading problem. The standard b-value DICOM Tag is used,  but different from GE and SIEMENS, the b-value was encoded as double floating point.

Investigating DICOM Rescale\Intercept problem with the Philips diffusion volume.

Finished ABT subtest for the PET workflow:  a 220 timepoint PET volume is loaded, registered to MPRAGE.  Resulting registered volume is average and compared to the stored standard.

Developed a support function for Automatic Build test to compare results of operation involving Real-valued volumes.  Up to now ABT was using IM format for checking the correctness of the computations.  But .IM is not suitable for comparing results with Real-valued data.

Implemented a feature so that during Automatic Build Test every tested dialog button can be pressed programmatically (i.e. not manually).  This way all the dialogs of all tested workflows would have a nominal testing, i.e. open with default parameters and press OK.

Finished PET workflow ABT subtest that loads a 20-timepoint PET Dicom, finds the dynamic average for all the voxels and compares it to the stored standard.

Working on the Automatic Build Test subtest to test current full PET processing workflow....

Three subtests are planned:

1.   Load PET Dicom dataset.  Use Dynamic averaging function.
2.  Load and register PET dataset to the anatomy MPRAGE dataset
3.  Run MRTM models

Released an unofficial update Build 125E to include some urgent fixes.

Enabled the registration of PET volume to MPRAGE anatomy using the URAL registration measure (and its valiants LIN and LOG).   Matching seems somewhat better than Mutual Info.

Fixed a major issue affecting 3D\4D registrations:  When texture gradient detector is performed the result is a 3D vector with Real (float double) components. For the purpose of performance and using SIMD, the vector components are converted into the int16 components.  To preserve the accuracy the real values are first scale (by 10,100,1000 etc) to keep precision and than rounded.  Those coefficients were fixed in all previous builds prior Build 126. It all seemed to work, i.e. registration results were good.
While enabling the URAL registration measures for Real-valued volumes (such as PET),  result was poor.
When investigated it was discovered that the URAL measure is accumulated within one row into the INT32 collector.  Its was frequently overflowing due to the the above ->int16 conversion.  Solution was to replace the accumulator with INT64 and also come up with the new scaling conversion. It all cleared with ABT and new PET tests for 3D\4D registration.

Screenshot:  PET to PET (two different timepoints) registered with URAL measure.

Enabled 3D registration using URAl vector measures for Real-valued volumes such as PET.

Fixed  the  3D Texture Gradient function to work properly on the Real-value volumes such as PET.

Fixed  the  3D Edge Detector function to work properly on the Real-value volumes such as PET.

Fixed error:  3D registration would not initialize for two Real-valued 3D volumes, such as PET.

Compressed Volume->Compressed Volume  copy RECT3D function - added an additional and more efficient level of the parallelism.  This is a starting operation in all 4D registration and can be quite lengthy.

Fixed an Initialization error in 4D registration dialog box.

Fixed defect:   4D registration would not initialize for 4D Real-valued volumes such as PET.
Presently, 4D registration with Mutual Info was tested on 4D PET volume.

Released unofficial "quick-fix" Build 125D to Henry.

Automatic Build Test problem:  in Dynamic Framework subtest failure on Ktrans-Ve model. Investigating...

Defect: crash in T2\T2* mapping model for the 4D abdominal volume provided by Henry.

Fixed: the problem was in incorrect handling of timing data, when the Dynamic Component is different from TIME (ECHO in this case).

Automatic Build substest for "Registration with Initial transform" is crashing.
Fixed an intermittent error in the core Registration procedure, when registration image padding was calculated based on the URAL radius, while that radius was undefined due to the different registration measure used (i.e. non-URAL, f.e. MI etc)

Validating the complete  Automatic Build Test for the Build 126...

Code work to simplify and reduce the parameter setting in several dozens of 3D\4D Registration scenarios.

After running the ABT (automatic build test) realized that registration examples using URAL measure all started to fail due to some kind of ripple effect. Investigating and fixing it...

Defect was in the "Final reslice Target-to-Source" option for all registration Measures, now fixed.