DBP:Harvard:Software:Testing:DTI:Slicer

• DTMRI vs. DTMRI_Matlab:
  • Table
  • Updated table (12/13/05): Media:DTMRI_comparison.xls
• DTMRI vs LSDI (ADC/Trace):
  • Table
  • Updated table (12/13/05): Media:ADC.xls
• DTMRI vs. LSDI (FAI):
  • Table
  • Updated table (12/13/05): Media:FAI.xls

11/15/05 More comparison of LSDIrecon vs. DTMRI

• Table with new testing data here
• Notes:
  • Using Matlab scripts to compute this data appears to be safe, except for:
    1. LSDIrecon DICOM files
    2. Marc: calculating Mode with Matlab?
  • ADC vs. Trace: Why the 251X difference?
    • NRRD headers give same results as DTMRI
  • Can we use Matlab scripts on DTMRI scalar measures for our studies?
    • Would this mean not using LSDIrecon?

9/28/05 Comparison of LSDIrecon vs. DTMRI

• Trace values are very small compared to ADC (may be a Scale Factor issue)
  • We have tried testing this using NRRD header data (as suggested), but we need Gordon to confirm that we are doing it correctly. Next week...
• The "Include Zeros" button in slicer_lmi DOES work correctly.
  • FAI and trace values are NOT different when including zeros.
    • Does this mean there are no actual "zero" values?
  • Now that the button is there, do we want to keep including zeros?
    • Currently the default is set to include zeros in slicer_lmi, but the function does not exist in slicer24, etc.

9/21/05 Comparison of LSDIrecon vs. DTMRI, using FA & ADC/trace as test variables. Also a comparison of the "include zeros" functionality that Steve included.

1. Differences in FA are always <0.25% (usually much smaller)
2. Raul removed the Scale Factor, which seems to make Trace values very small compared to ADC
   • BUT DTMRI trace values are still consistent (in relative terms) with LSDIrecon ADC, i.e., always 250X smaller
3. Including zeros (Steve's added button) did not make a difference. Does this mean 1) there is no difference, or 2) the "include zeros" button does not work?

6/14/05: Comparison of LSDIrecon output (FA and Diffusivity) to DTMRI scalar volumes

1. Differences in FA are small (typically less than 0.25%), but why are there differences?
2. Differences in ADC/Trace exist, probably due to the Scale Factor. Mark has emailed Raul about this.
3. Can other measures be compared between LSDIrecon and DTMRI?
   • LSDIrecon outputs: AAI, ADC, DWI, FAI, FIL, L1x, L1y, L1z, RAI, T2W, VAI
   • See a description of Stephan's LSDIrecon program (from /projects/schiz/software/bin/LSDIrecon.HOWTO, 6/15/05, but is this an older version of the file, 2003?)
4. Note: Diffusion images were sagittally acquired, and the labelmap used was Mark's manual corpus callosum drawings

6/1/05: Sylvain and Mark met with Stephan a couple weeks ago to discuss the slice corruption problem with the lsdi2dwi script and DICOM cases (see notes from 5/13/05). Stephan discovered the source of the problem: the swab function behaves slightly differently under Linux and SUN.

1. Stephan fixed the problem
2. Sylvain implemented the code in the lsdi2dwi script
3. Mark confirmed that the new script is reliable.
   • It is now safe to use the lsdi2dwi-all.pl script on DICOM cases.

5/20/05: Below is the table confirming that Gordon's matlab scripts accurately differentiate Fermi-filtered from non-filtered cases.

• NOTE: T2W or DWI must be available to use this method.

Notes for this table:

1. "Fermi filter" column was determined from Gordon's images, and confirmed using the test data below.
2. Note that for highlighted cases, data in the "original .pic file" column was NOT filtered, and results are different from data using the Fermi-filtered lsdi2dwi script.

5/18/05: Gordon's results

5/17/05: Erin and Mark found out that a majority of cases were potentially reconstructed before Stephan's LSDIrecon v1.3 was implemented, so they are working with Gordon on a method to retroactively determine whether cases were reconstructed with a Fermi filter.

• Question for Stephan: What were the exact dates that changes in LSDIrecon were created?

5/13/05: Sylvain modified the lsdi2dwi-all.pl script, and Erin & Mark compared the output of this script with the original versions of the LSDIrecon scripts. The output table is below.

Important notes/summary:

1. The lsdi2dwi-all.pl script is (as of now) reliable for GENESIS CASES ONLY
2. Two problems were discovered regarding the use of lsdi2dwi-all.pl script with DICOM cases:
   • Output is in Little Endian format
   • Several slices in each case become corrupt, leading to slight changes in FA/ADC values (see highlighted cases below)
3. We will once again recruit help from Stephan to fix the problem of using this new script with DICOM cases.
4. For now, it is ok to use the lsdi2dwi-all.pl script WITH GENESIS CASES ONLY. DO NOT ANALYZE DICOM CASES UNTIL FURTHER NOTICE!

In the table below:

• CB = Cingulum Bundle
• caseD00760 (highlighted) is DICOM

5/11/05: Mark and Marek met with Stephan Meier and identified one source of the differences found in the LSDIrecon scripts.

• Data in /projects/schiz/diff/chronic/males was obtained using either LSDIrecon version 1.2 or 1.3, which differ in the use of a Fermi filter in the latter version
• It is unclear which cases were run with and without the Fermi filter, and it will be difficult to determine this in most cases
• It has been proposed that anyone using data in /projects/schiz/diff/chronic/males in the past 2 years should re-analyze their data

5/11/05: Stephan also brought up the issue of a threshold used in his LSDIrecon scripts, which could also change numbers slightly.

The following is a copy of the parameters in the LSDIrecon_par file:

• 1.3 # DO NOT CHANGE (LSDIrecon version)
• 1 # parameter file: 0=long, 1=short for LSDI, 2=short for other
• 1 # Fermi filter in read direction: 0=off, 1=on
• 1 # correct missing lines: 0=no, 1=normal, 2=strong
• 2 # correct eddy current distortion: 0=no, 1=yes, 2=yes incl. NEX
• -10 # clipping threshold (<0 noise multiple, =0 none, >0 absolute)
• 1000 # b-value for reconstructed DWI, 0=maximum b-value
• 1 0.2 2.2 0.4 -0.3 0.5 0.2 # fit order (order: >0 constant amplitudes, <0 tensor amplitudes), min ADC1, max ADC1, step ADC1, min ADC2, max ADC2, step ADC2
• 1 # data type: 0=Signa 5.4; 1=5.6, 5.7, LX
• 1 # output: 0=nothing (for testing), 1=basic, 2=basic with header, 3=everything, 4=everything with header
• ./ # Path for output files, use ./ for local directory
• test # Prefix for output file names
• ./ # Path for input files, use ./ for local directory
• 0 # nr of images to offset during autoloading (usually 0)

4/25/05: The DTMRI module in slicer_lmi now produces more reliable results (see below).

• A DICOM case is now included (caseD00760)
• DTMRI now gives results similar to the lsdi2dwi-all.pl script (but not exactly the same in all cases?)
• We must now determine how to use the "Scale Factor" to report correct ADC/Trace values
• The problem still exists of why the new lsdi2dwi-all.pl gives slightly different results from Stephan's old lsdirecon script (located in /projects/schiz/software/bin)

4/20/05: DTI data was obtained using 3 different "background" data sets:

• 1) Stephan's "old" LSDIrecon scripts
• 2) the new lsdi2dwi-all.pl script created by Sylvain and Stephan
• 3) tensor data, by converting DWI data and creating Scalar volumes, then casting float --> short to mask using labelmaps

Notes:

• labelmaps were of the cingulum bundle used in Marek's study
• both cases shown were obtained in Genesis format
  • caseD2 acquired 1/9/00
  • caseD0667 acquired 7/19/03