While "Snap-to-Point" feature works quite well on femur, there are areas where local features are not sufficient to reliably detect the desired location

Problematic boundary for "Snap-to-Point"

Typical structure of the edges.  Local structure is not sufficient to reconstruct a good snapping location.

To solve this problem I propose to include additional mechanisms, see figure below.

"Snap-to-point" would simply snap to a stronger edge within the Snap apperture. This would be incorrect. Instead taking the previous state of the cardinal spline into account, we search for a better snap location.  Snap-to Curve would inregrate the edges strength all along the last segment, so the yellow path would be invalidated. And instead more reliable path chosen.

FireVoxel Build 210A is released.

Implemented the functionality to load real valued 4D Analyze files (*.time+ array of .IMG files)

First runs of new MagPoly tool on the femur:

First user performs several clicks on perceived contour of the femur. Computer calculates the edge pattern next to the click and selects a "snapping" location setting an anchor point.
Two important points:

a) Even if snapping did not work properly - operator proceeds with the next points (there will be a cleanup stage later).
b) Work is performed on a full Zoomout, so User does not have to Zoom In and back and scroll!
c) In between the anchor points, a cardinal spline with application-specific tension is used.
d) 10-voxel snapping radius was used.

Each anchor point corresponds to a single click

In the followup stage. User returns and drags some anchors to improve their positions. In this example about 6 anchors were corrected.

After dragging 6 anchors.

MagPoly tool: a simple interface to extend the functionality of the Polygon. It is purely 2D tool. No 3D extension is foreseen at the moment.



Tool is activated with the present Polygon tool. Then behavior changes according to dialog parameters.

|x|  UseMagPoly - turns the tool on.  Off = regular polyline.

"MagPoly radius":  the cursor changes to a semitransparent circle (similar to the paintbrush cursor). Radius specify the snapping seaarch radius - this way user has some feedback how far she should click.

|x|  Precompute slice edges (for slow computers).  - at the moment I plan to compute edges each time user clicks within the apperture specified by the "MagPoly radius".  Hopefully it will fill instantaneous on the mainstream computer.  Otherwise, with the tool on each time the slice would change, the internal edge layer will be computed.

"Edge Parameters" - parameters identical to the Texture Edge detector.


Additional parameters will appear later on.  F.e. more advance version would snap the curve between the user-specified Anchor points.

Implemented "ROI>Concentric Ring profile" function with the result being copied to the clipboard.

Idea is to capture the spatial characteristics of the tumor by detecting Texture-sensitive edges within the tumor at different scales and different apperture weights.

EMD texture edges, Scale=3, Apperture=constant

 Here are the corresponding Edge strength distribution histograms:  They have a widely varying StdDev/Mean and so on.

We can also Build the "Concentric Profiles" of Edge Strength distribution ( in corresponding order)

"Tumor blob analysis"

Tumor is thresholded.

Dark blob info is available

"Tumor ring profile" trials:

Mean=Blue, StdDev=red,  1 vox step -> inwards

Full Resolution Femur sample:
Run the registration Using URAL, default parameters, Subsample=8,with Finetune=1.
Timings:
Primitive generation:   3 min
AutoFocus:                  50 sec
FineTune:                     25 sec

Result is below.  I don't see a need to change anything about the registration FineTune.
MI problem has to be investigated.

Femur Full-resolution registration sample:

URAL registration works well with the present default parameters. Below is the result for the AutoFocus stage (no Finetune)

Using URAL, AutoFocus-only, no Finetune

However, for the MutualInfo this stage fails with the result grossly misplaced.

FGT segmentation: Implemented a hybrid FAT thresholding method based on Z-alg for when histogram has 2 distinct peaks. New criteria was developed for the absence of such peaks. In that case the PVV segmentation is invoked.
Processed all the 16 "challenge" cases with good results.

Developed completely new application-oriented AirThresholding method. It is based on Z-alg idea.
Validating it on the failed breast FGT samples where selecting across-the-board threshold = 60 did not work.  Specifically, in REIC cases the proper thresholds were substantially higher :  {67,108}
in KELL cases the proper threshold were substantially lower {17,15}

REIC_R

KELL_R

KELL_L

REIC_L

Added the {BiGauss_PVV,BiLaplace_PVV} methods to the menu of Histogram segmentation methods in
a) RoiStats3D dialog
b) "ROI>Split ROI with threshold" menu

Improved the performance of the "Interval Arithmetic" library. This is used in all Histogram segmentations and dynamic modelling.

Developed new thresholding methods  "Bi{Gaussian,Laplacian} with PVV correction".
These methods work on the idea that the histogram segment between 2 peaks is unreliable due to the Partial Volume Voxels (PVV) and should be completely disregarded.  Intuition here is how human thresholds complex fused histograms, just by looking on the side tails of the distribution.

Tested on the difficult case ABBE-R. With the good result.

Histogram modelling

Final segmentation

problematic case ABBE-R
seems the BiLaplacian split found a wrong threshold:

ABBE-R.  BiLaplacian split.  Residual = 2549

However, when we re-segment the BiGaussian model, we obtain much lower residual.

BiGaussian fit.  Residual = 645

ABBE-R:   Split using BiGaussian

However, for Phantoms the BiLaplacian was much better than Gaussian at least for most cases and all phantoms were processed using exclusively BiLaplacian.


Proposal:  We evaluate histogram fit using both {BiLaplace,BiGauss}.  We choose the method that provides smaller fitting residual.

Determine the Air threshold in Breat FGT images:
We inspect the histogram that has a gross-dominant peak around the Signal=0. We fine first evident valley and assign the valley bottom to be the AirThreshold.

New workflow on case ABBE_L:

result of the segmentation:

Testing new FGT workflow on human data:
BiCal + BiLaplace provide excellent result. Exactly the parameters  from Phantom processing
were used

BiCal followed by BiLaplace

Breast FGT phantom workflow:  implemented completely new workflow  BiCal+BiLaplace+MaxCC+EdgeWaveML.  Processed all 24 cases with very good result matching the physical measurement.

Some voxels are marked as FGT, but should be they included at all?  They could be fairly easily excluded by demanding that the FGT+FAT is a Connected object and does not consist of the several
disconnected components. This should be true for Phantom and Human data?

"EdgeWave MultiLabel - Boundary Only" variant was implemented.  Only the boundary voxels of the total region are modified by Multilabel Peel\Grow.
Provides slightly different result

FGT=187.8 cm3, FAT = 944, Total = 1131.8, Fgt%=16.5%

Implemented new algorithm "EdgeWave MultiLabel". This is a general algorithm: given an initial segmentation of the region with multiple labels it applies EdgeWave morphological criteria to correct these regions while completely covering the total region.

This was applied to p1s0_L example that had issues with partial volume voxels being misclassified as FGT.

This algorithm contains both internal and external morphological competition.  So additional algorithm "EdgeWave ML Boundary" will be attempted.

FGT: 180.23, FAT=950, total=1130.23,  FGT percentage=15.9%

Implemented the original variant of the FGT breast workflow.  It consists of BiCal, Bimodal Laplace thresholding.  While providing great (+-4% ground truth) result for some images, other images are problematic. This requires to develop an additional Morphological module, as none of the existing variants of EdgeWave don't seem to fit.

This is the "fused peak" case that Henry sent earlier.  As far as I understand this was the case "p4s0.fvx"-RIGHT (DICOM5)

Again BiCal was applied with the exact parameters as with other cases. Then BiLaplacian with exactly same parameters too.  We see a great histogram split as a result.

Initial fused peaks case resolved.

Corresponding segmentation (top and bottom slices)

FGT processing update:

1. Preliminary results on 4 phantoms with about 200 manual seeds each indicate x4-5 times improvement of BiCal over N3.

2. After fixing the BiLaplace fitting procedure, the phantom histogram splits start to look great. But no great conclusions till the whole set is done.  See below (case p1s1_R.fvx)

Histogram fitting using Bimodal Laplacian (after BiCal)

Understanding breast phantom data:
Provided phantom data should be
a) Thresholded to get the Breast ROI (instead of manual ROI in patient data).
b) L&R breast separated
c) Run NU {N3,Bical}
d) Run Histogram thresholding {Otsu,BiGauss,BiLaplace}
e) provide the results of measurement in cm^3
f) best method matching the ground truth will be selected

BiModal histogram segmentation within the RoiStats3D dialog box.

a) Added the separate "Recompute" button to simplify the processing
b) Currently selected\displayed histogram binning is supplied and used as the distribution for the BiModal modelling.
c) Corrected the Model Curve overlay to be consistent with the change in binning specified by the User.
d) Eliminated the "BiGauss model"  only leaving the "BiGauss with Ratio model".  Just set the very high peak ratio (>10) to get result of the previous models.
e) Fitting Thresholds\Residual are now displayed in the "Threshold" status line.

BiGauss model (residual 57.7)

BiLaplacian model (residual 86.6)

Enabled a correct progress meter inside the Breast-FGT "compounding workflow".

Developed an adaptive function to separate breast image into L&R.  This is to compensate for 2 possible operator related errors:  initial acquisition centering and positioning of the cropping ROI.

BiGauss Histogram modelling (on the ventricles image):
Observation is that the "With Ratio" variant executes much faster than "Without Ratio" variant.
But then on this particular sample  the second one seems to have much better result.

Bimodal Gauss WITH Ratio

Bimodal Gauss (without Ratio)

BiGauss Histogram (without ratio) modelling:  Added the progress bar.

Improved Automatic DICOM loading function (does not require user input). Now function may be supplied with the file folder name. Before it required only the image file name of the actual dicom file.

FireVoxel Build 209 is released.

 1. Fixed the "IM file crash problem" (equal sign was missing):  an appropriate error message would be displayed now.

2. IVIM segmented model:

a) added parameter "Large B-value start" with the default value=200.  User can now change that value.

b) Increased the allowed ranges for Dp=[0,1],  Dt=[0,0.01]

IVIM basic model:

 Increased the allowed ranges for Dp=[0,1],  Dt=[0,0.01]

3. On Mouse Move - Display voxel information:  fixed a defect when the Active Layer is disabled, so moving the mouse did not produce any information in the status line.

Now if the Active is disabled, we choose the First enabled layer to display the voxel info.

4. Synchronize Multiprojection Cursor (mode=ON) functionality:  if there is a single View\Projection of the volume is visible the Crosshair cursor would not be visible.

5.  BiCal on Sodium images: fixed the defect when iterations of BiCal were producing a "better" result. This was due to the custom scheme of processing the REAL images, when internally they were converted to the 15 bit images and this was conflicting with the internal truncation  

Verified "Registration by DICOM-tags" on the PET-MR sample provided.  Matching quality seems to be good.  Since torso is included, it is unlikely any other registration could\should be used.

Fixed the "IM file crash problem" (equal sign was missing):  an appropriate error message would be displayed now.

IVIM segmented model:
a) added parameter "Large B-value start" with the default value=200.  User can now change that value.
b) Increased the allowed ranges for Dp=[0,1],  Dt=[0,0.01]

IVIM basic model:
 Increased the allowed ranges for Dp=[0,1],  Dt=[0,0.01]

"IVIM-Segmented"  model:
We try to investigate a suspicious peak in the histogram of the fitted Dp-parameter.  This peak occurs at Dp=0.03.  In the screenshot below one of such voxels was identified and processed individually.

Green-data curve.  Black - fitted curve.

For finding the Dp component we run the Global Optimization with parameter Dp be restrained to the [0,0.03] interval.  So optimal solution is found to be on the borders of that interval, i.e. Dp= 0.03.
Many such voxels within the ROI contiribute to the "suspicious peak".

The problem does not seem to be restricted to the cases of air/tissue partial voxels.
My current guess that this is common to the voxels where signal curve is not monotonous.

Here is anther case with Dp=0.03

Next experiment, I vastly increased the Dp-constraint interval to [0,1] and ran it on the same voxel.  Interestingly the result has settled on Dp=0.67 with 3 times lower residual.  This indicates that we have found an actual optimum.

Fit with Dp=[0,1] allowence. This settled at Dp=0.67

CONCLUSION:  seems like the presence of the  Dp- histogram tail peak is due to the restriction of the Dp=[0,0.03] during the optimization.  I will be relaxing this restriction to Dp=[0,1].
My best guess that the "presence of the histogram tail peak" is not related or causing the difference of the results with other software.  I use the Global Optimization which is very different from Local Optimization used in other software.  To my best knowledge the FireVoxel fits are valid and excellent (judging by the residual). So my best guess is that "other software" is not fitting well enough.

Here is the histogram of resulting Dp values with  Dp=[0,1].  Previous strong peak at Dp=0.03 is gone. My best guess: these high values due to some voxels having strongly monoexponential signal
curve.

We can get to the bottom of it if I send a single Signal Curve, along with FireVoxel-calculated parameters and residual.  Then curve can be processed in "other software", residuals compared and so on.

On Mouse Move - Display voxel information:  fixed a defect when the Active Layer is disabled, so moving the mouse did not produce any information in the status line.
Now if the Active is disabled, we choose the First enabled layer to display the voxel info.

Synchronize Multiprojection Cursor (mode=ON) functionality:  if there is a single View\Projection of the volume is visible the Crosshair cursor would not be visible.

BuiCal nonuniformity correction over the set of 12 "Sodium"-images.

To precisely evaluate the effect of the correction, 15-20 seeds were precisely and manually constructed on each of the cases.  All of the seeds were positioned inside of the ventricles as far as possible from the edge of ventricles and corresponding partial volume voxels.

Non-uniformity is defined as the StdDev of the average seed signal taken over all the seeds.

Seeds are in Red on one of the slices 

We tried to find set of parameters that worked well on every single image.

See the Table of results below for each case  

1st column:  Case name

2nd Column:  original non-uniformity (StdDev over seeds)

3rd Column: processed non-uniformity (StdDev over seeds)

5th column relative improvement of the correction =NUbefore/NUafter-1

Interestingly the results for N-cases where 10 times weaker than the results on "Non-N cases", 6% and 61% improvement correspondingly.

N3 correction was attempted but NU only worsened with all the parameter combinations we tried.

Overall these images are quite challenging for Non-uniformity correction, due to the low resolution and the ventricles being only 3-4 voxels thick.  Note the original .NII files contained the resolution of (1,1,1)mm but this does not seem to be the case,

BiCal on Sodium images: fixed the defect when iterations of BiCal were producing a "better" result.
This was due to the custom scheme of processing the REAL images, when internally they were converted to the 15 bit images and this was conflicting with the internal truncation mechanism after the correction.

Sodium image for testing the BiCal non-uniformity correction.  NII file resolution is listed at 1mm^3 (dimensions are 128x128x128) but it seems incorrect.  User desire the signal in ventricals (see the seeds) to be uniform.

Build 208   is released.

1. IDIF function: Fixed the "invalid rectangle" defect exhibited on P1.fvx.  This was due to the inflated initial rectangle to be outside the image bounds.

2. Implemented an order-of-magnitude CountBits( PBYTE Array,int Start,int End) function which is frequently used throughout.

3. Dynamic Modelling (models {0,3}): fixed the user defined parameters for starting dynamic index to 1 (from 0).

4. Dynamic Modelling framework fix:  When PET image is processed with "Process ALL", i.e. without the ROI, the dialog box was popping up asking to specify T1.

5. Dynamic Modelling:  added a test for Tissue Concentration if the Modality=PET. In this case the Concentration Method is always set to "SIGNAL" and user is warned.  This is done in addition to an earlier implemented check for the Input Function in PET.

6. Dynamic Model "Input function correlation": Added a proper handling of the Tissue concentration. Before it was "Signal only" and created inconsistencies for the user.

7. Dialog Layer Operations: fixed crash defect related to the Entity Layer upgrade to 22 layers in the previous builds.

8. Fixed the error in reslicing binary volumes (ROIs).  Old procedure was always assuming the FillVoid or Background value is always '0' but this is not correct always.  This was detected in "Reslice Target to Source" option during the Femur registration.

9. Dynamic Modelling:  Input Function correlation: corrected the situation when the divisor n*XX-X*X or n*YY-Y*Y is close to zero.

10. Dedicated menu item was added to the workflow menu with the parameters suited for processing.

11. LayerControl dialog: updated behavior so that the Name editing requires Doubleclick, while "ROI advance to midslice" requires "Ctrl+Doubleclick".

IDIF function: Fixed the "invalid rectangle" defect exhibited on P1.fvx.  This was due to the inflated initial rectangle to be outside the image bounds.

Implemented an order-of-magnitude CountBits( PBYTE Array,int Start,int End) function which is frequently used throughout.

Dynamic Modelling (models {0,3}): fixed the user defined parameters for starting dynamic index to 1 (from 0).

Dynamic Modelling framework fix:  When PET image is processed with "Process ALL", i.e. without the ROI, the dialog box was popping up asking to specify T1.

Dynamic Modelling:  added a test for Tissue Concentration if the Modality=PET. In this case the Concentration Method is always set to "SIGNAL" and user is warned.  This is done in addition to an earlier implemented check for the Input Function in PET.

Dynamic Model "Input function correlation": Added a proper handling of the Tissue concentration. Before it was "Signal only" and created inconsistencies for the user.

ABT :  added a subtest for basic functioning of the Layer Operations dialog.

Dialog Layer Operations: fixed crash defect related to the Entity Layer upgrade to 22 layers in the previous builds.