Fascicle Segmentation


Jump to: navigation, search

This page has been superceded by the more recent work using Automated Fascicle Quantification.

There is an original publication describing that method.

Tract profiles of white matter properties: automating fiber-tract quantification. Jason D. Yeatman, Robert F. Dougherty, Nathaniel J. Myall, Brian A. Wandell, Heidi M. Feldman (2012). PLoS ONE 7(11): e49790. doi:10.1371/journal.pone.0049790

This page was written by Davie Yoon and it illustrates the kind of careful hand work that was necessary to perform tract segmentation in the early years.

Related pages are MrDiffusion, LiFE, and MRtrix.


[edit] Quench Procedure

Both manual plane-setting procedure and the Mori atlas procedure have significant problems with regard to tracking. The manual plane-setting procedure only tracked fibers that intersected both planes, which were manually set and drawn following protocols developed by Wakana (2007). However, this tracking method often missed many shorter fibers that do not extend all the way from one plane to the other, which is a significant problem since fibers that are likely related to face processing (for example, connecting to the FFA) may be shorter and will thus be overlooked. Manually setting ROIs is also a time-consuming process.

Classifying tracts according to the Mori Atlas (further explanation below) is also problematic. Preliminary results did show hemispheric and age-related differences in fiber tracts, but these were due to intrinsic bias in the normalized brains, not due to actual systematic differences.

The Quench procedure attempts to bridge some of these problems. This procedure uses a single ROI as a tracking criteria, and then the resulting fiber paths are manually cleaned in Quench. Currently, we only have protocol for tracking the ILF.

[edit] Using Quench

  • To open: type Quench in terminal window. It's best to cd to subject's directory first to ease file access.
  • To manipulate slices
    • Change selected slice by typing a for axial, s for sagittal, and c for coronal.
    • Hide a slice by typing shift + a/s/c
    • Scroll through slices with mouse wheel
    • Right click to zoom, left click to change angle
    • Shift + left click to move images.
  • To hide/show pathways: either click on colored numbers on lower left corner of screen, or type number
  • Touch tool: ctrl+left click to draw across certain fibers. If + is selected, this will add these fibers to whatever group number is selected. If - is selected, this will delete these fibers.
  • Surface tool: Switches to surface view. May take about a minute the first time used. Ctrl+left click to select certain fibers by drawing around them. Adding/deleting fibers as above.

[edit] ILF Protocol

  1. Open a terminal window and cd to subject's directory. Type Quench.
  2. In Quench, File--> Load Volume --> t1. Select t1 nifti to open.
  3. File --> Load Pathways --> dti30_trilin/fibers/wakana2007_quench. Load ILF fibers.
  4. Start anterior and scroll to posterior, looking for slice where the fibers first diverge into two branches. This should occur around where the temporal lobe first completely separates from occipital lobe.
  5. Check in surface view. Use surface tool to outline entire temporal lobe and transfer to new color group. Hide first color group, leaving only selected fibers in view.
  6. Remove any fibers through parahippocampal gyrus by using touch tool to transfer them to a new, "excluded" color. The will be medial to the main body of fibers, directly in white matter arm. Not all fasisculus will have fibers in parahippocampal gyrus.
  7. Remove any fibers that go parietally, past the parieto-occipito sulcus. Add these to "excluded" color.
  8. Remove problematic fibers that are likely artifacts: fibers that loop back on themselves, individual fibers that go off on themselves, and descending arms that are perpendicular to main body.
  9. Save fibers: Make sure only one color group is visible at a time. Save as RH_ILF_manual.pdb or RH_ILF_excluded.pdb, as appropriate. If using an ROI other than the Wakana-defined ROi, save as RH_ILF_manual_FFA#.pdb or RH_ILF_excluded_FFA#.pdb.

[edit] Drawing Alternate Planes

The original Wakana-defined ROI may be missing shorter fibers that pass through the FFA. Drawing a plane anterior to the FFA may help catch these fibers.

  1. In Matlab, cd subject's dti30_trilin folder.
  2. Type: dtifiberui dt6.
  3. File-->load ROI --> subject's directory --> dti30 --> ROIs --> Functional. Load RFFA_MBvAClO_p3d_gm1.mat.
  4. ROI--> Find current ROI --> Note Y coordinate.
  5. Move cursor 10 mm anterior (add 10). Note Y coordinate.
  6. Draw ROI that encompasses entire hemisphere (follow procedure outline below for ILF ROI 1).
  7. File --> Load --> thomas2009. Load RH_ILF_ROI1.
  8. ROI --> Find current ROI --> Note Y coordinate. Not difference between RH_ILF_ROI1 and the ROI you just drew.
  9. Save your ROI in Wakana2007 folder as RH_FFA_plane# (# = number mm moved anterior).

[edit] Plane Setting via Manual ROIs

The following ROI segmentation protocols for fascicle segmentation is derived from Wakana et al (2007). This paper builds on previous work from Mori (2002), but focuses on the reproducibility of tracking results. As compared to the ROIs defined by Mori, Wakana's ROIs are much larger to limit the effect of variations in individual brains as well as in different raters. Consequently, more fiber cleaning may be necessary after tracking is complete. However, while the reproducibility ratings of Mori's ROIs are unknown, Wakana's ROIs have been tested with three different raters within their institution as well as three outside raters from different institutions.

[edit] General Tips, Instructions, Help

Unless otherwise noted, all ROIs are drawn on the t1 image on a coronal slice.

[edit] Opening dtiFiberUI and Loading Files

  • In Matlab (version R2008a), cd to Z:\Kids\dti\thomas_2009_analysis\person working on. You can tab complete for each step to prevent typing errors. If you accidentally go to the wrong folder use cd .. to move up one folder. To see the contents of a folder, type ls.
  • Find the dt6.mat file. It should be under the person's dti30 folder. If not, take note and move on to next person.
  • Type: dtiFiberUI dt6. Once you've opened someone's slices, DO NOT cd anywhere. If you need to see contents of a folder, just use ls.

[edit] Using dtiFiberUI

  • Loading ROIs: File-->ROIs-->Load many ROIs
  • Saving ROIs: After each ROI is drawn, first rename ROI by clicking the edit button next to the Current ROI menu in the lower left hand corner. Use this naming convention: RH/LH_fibertract_ROI1/2. Next click File-->ROIs-->Save Current ROI. Save the ROI in the thomas2009 folder under the subject's ROI folder.
  • Drawing ROIs: ROIs-->New(polygon)-->Y Image. Click to set points for ROI, right click to close. Double click inside ROI to create red mask. Follow the gray matter closely, but don't both cutting out gryri. Be sure not to include CSF between brain and skull. Also, sometimes the dura is visible as a thin gray line surrounding the brain. The dura is not part of the gray matter, so it should not be included if it strays away from the hemispheres. If you are uncertain whether an area is still gray matter, you can click on it and check against other slices.
  • Editing ROIs: ROIs-->Edit Shape-->Add/Delete Polygon-->Y Image. Click to set points, right click to close. Double click inside ROI to add/delete new area. It can be tricky to modify an ROI because the program will only mask the voxels inside the polygon, not on or partially across the polygon. It may be easier delete the ROI (ROI-->delete some ROIs) and redraw entire ROI. Note that deleting an ROI only removes it from the list of loaded ROIs and does not actually delete a saved file.

[edit] Order of ROI Drawing

It is very easy to lose track of which ROI is being drawn. To maintain consistency and prevent errors, please keep these guidelines in mind:

  1. Draw ROIs in this order: ILF ROI 1 and 2, IFOF ROI 1 and 2, UNC ROI 2, F major, F minor.
  2. Complete all ROIs on right hemisphere before doing left hemisphere.
  3. After drawing each ROI, be sure to rename, save (using correct convention), and update spreadsheets accordingly. One sheet should contain information that will be needed for later ROIs as well as any notes about alignment, artifacts, venticles, etc. Use the second sheet to keep track of which ROIs on which hemisphere have already been drawn.
  4. After completing each hemisphere, make sure there are seven different ROIs.

[edit] Drawing Guide

  1. Draw all ROIs on t1 image.
  2. Start at the most superior medial point.
  3. Draw down (inferior), and then move laterally and up to complete ROI.
  4. On right hemisphere, this means that ROIs are drawn counter-clockwise, while ROIs on the left hemisphere are drawn clockwise.

[edit] Alignment Checking

There may be errors in the alignment of the dti data(b0, FA, MD, vector RGB) and the anatomical data (t1). Only the FA and vector RGB are from the same images (b0). One of the easiest ways of checking alignment is to compare the corpus callosum on the T1 and diffusion images (b0, FA, MD, RGB). The images may be properly aligned on most, but not all of the borders, suggesting the image was stretched/contracted in that direction. If you find an error, take note. Thus far, it seems that the inferior border is very often mis-aligned.

If there is a discrepancy between the t1 and the diffusion images (b0,FA,MD,RGB), follow the t1 image to draw ROI as long as doing so will not intrude into other brain areas. For example, it's okay if the ROI exceeds the brain mask on the other filters. However, if drawing the ROI on the t1 filter will intrude onto other brain areas (i.e. other lobes, extra gray matter), take note and talk to Davie. DO NOT change criteria to fit other brain filters.

Alignment should be checked for each ROI drawn. It can be checked by briefly flipping to one of the diffusion images, whichever one seems most informative for the border you want to check, and then comparing the superior, lateral, inferior, and medial boundaries. Add notes in the Excel sheet about the nature of the deformation by estimating number of missing/additional voxels and location of distortion (along entire border or single area). Highlight notes describing missing data in excel spread sheet. Include which ROI and hemisphere distortion is found.

[edit] ILF definition protocol

The ILF protocol has the lowest reproducibility rating for different raters (see Wakana et al, 2007 Table 1).

[edit] ROI 1: Occipital Lobe

  1. Begin in middle sagittal plane, using the RGB vector filter.
  2. Use +/- buttons to move laterally in sagittal view. ROI 1 is drawn on the most posterior edge of the cingulum, the green fiber wrapped around the C-shaped corpus callosum in the center of the brain. As you move more laterally, the cingulum will curve around and become blue at its vertical point. Choose the most posterior edge of the blue cingulum as the drawing slice.
    1. If needed, reference the MRI Atlas of Human White Matter, pages 225-231. The sagittal views clearly shows the cingulum and where it disappears, while the coronal and axial views provide an indication of how lateral you should be.
    2. Switch to t1 anatomical image. Make sure the slice is still close to corpus callosum.
    3. Draw polygon ROI delineating entire occipital lobe by following gray matter along edge.
      1. Rename and save ROI: LH/RH_ILF_ROI1.

      [edit] ROI 2: Temporal Lobe

      1. In t1 image, on the sagittal slice, click on the anterior commisure. This will appear as the bright dot at the base of the corpus callosum. Alternately, enter (0, 0, 0) as the current coordinates.
      2. Switch to coronal view. Scroll anterior (+ button) to where the frontal and temporal lobes separate. There should be a small line of CSF dividing the two instead of a white matter band. If there is a small uncertain area (i.e. mostly separate, but a small band of gray matter still connected), switch to MD filter. If the temporal lobe has completely separated, there should be a clear white line between the temporal and frontal lobes. Write down this slice.
      3. Follow gray matter around to delineate entire temporal lobe.
        1. Rename and save: LH/RH_ILF_ROI2

        [edit] IFOF definition protocol

        [edit] ROI 1: Occipital Lobe

        Note: This ROI is different from the ILF occipital ROI.

        1. Start on the same slice as ILF ROI 1. The easiest way to do this is to re-enter coordinates from the slice you previously drew on.
        2. On this slice in the sagittal view, choose the most posterior point on the parietal-occipital sulcus (POS). Choose point in middle of POS, not towards one of the borders. Save these new coordinates.
          POS sagittal.jpg
        3. ROI is drawn the coronal slice directly between these two slices: (1) ILF ROI 1 and (2) posterior point on POS. Average the Y coordinates to get the midpoint.
        4. Type in the averaged Y coordinate. While on the sagittal view, move cursor progressively inferior until the cursor reaches the POS. Now switch to the coronal view. Take care NOT to shift the y coordinate.
          POS coronal.jpg
        5. Linearly extend POS laterally to form superior boundary, then follow gray matter to delineate occipital lobe.
          IFOF ROI1.jpg
        6. Rename and save: LH/RH_IFOF_ROI1

        [edit] ROI 2: Frontal Lobe

        1. Switch to RGB vector filter. On a sagittal slice, identify the most anterior edge of the genu of the corpus callosum, which is the bright red fiber bundle in the center of the brain. Useful pages in the Atlas are 229-237. Check on t1. Switch to coronal view. Save these coordinates.
          1. This ROI will encompass the entire hemisphere. Begin at most medial, inferior point of the hemisphere and draw counter-clockwise following the edge of the gray matter. The medial border should extend vertically from the most superior point to the most inferior point of the hemisphere.
            IFOF ROI2.jpg
          2. Both the right and left hemisphere for this ROI will be draw on same slice.
          3. Save and rename.

          [edit] UNC definition protocol

          Both ROIs below are drawn on the most posterior coronal slice where the temporal lobe and frontal lobe are seperated.

          [edit] ROI 1: Temporal Lobe

          This ROI is identical to ILF ROI 2.

          [edit] ROI 2: Frontal Lobe

          1. Find ILF ROI2 in coronal slice. Switch to RGB vector filter.
          2. The Atlas provides good images of both tracts on pages 131-139.
          3. Next orient yourself on the coronal slice:
            • A little superior and medial to the inferior boundary of the frontal lobe, there should be a green fiber tract that is C-shaped (concave aspect is pointed medially towards the ventricles). This is, roughly speaking, the UNC. In the images below, the UNC is indicated by the white arrows.
            • There should be another green fiber tract bordering lateral side of the ventricles. This is the corticopontine tract (CPT) and forms the superior and medial border of the new ROI. In the images below, the CPT is indicated by the yellow arrows.
              UNC zoomed in t1.jpg
            • Switch to t1 image and identify same tracts again.
          4. Drawing ROI: On t1 image, define ROI. It should include entire UNC and gray matter between that and CPT. It should NOT include CPT. These borders can be difficult to see, so if needed, use the zoom button on the middle bottom of the screen to enlarge image.
            UNC ROI in t1.jpg

          [edit] Forceps Major definition protocol

          Both ROIs may be drawn on different slices, but follow the same protocol.

          1. For IFOF ROI 1, the slice containing the most posterior point of the parieto-occipital sulcus was defined. Return to the slice.
          2. Switch to coronal view. Delineate entire occipital lobe by following gray matter outline.
          3. The right and left ROIs may be on different slices, but if there is a huge difference, talk to Davie. Even on different slices, right and left ROIs might be different sizes and shapes.

          [edit] ROI 1: Right Occipital Lobe

          ROI location.jpg
          Fmajor ROI.jpg

          [edit] ROI 2: Left Occipital Lobe

          [edit] Forceps Minor definition protocol

          Both ROIs are drawn on same coronal slice.

          1. When you drew IFOF ROI 2, the slice with the anterior-most point of the genu of the corpus callosum was defined. Return to that slice.
          2. In the saggital view, identify the anterior-most point of the frontal lobe on a t1 image. Write down slice number.
          3. ROIs are drawn on the coronal slice on a t1 image directly in the between the above slices.

          [edit] ROI 1: Right Frontal Lobe

          Midpoint Fminor.jpg
          Fminor drawing.jpg

          [edit] ROI 2: Left Frontal Lobe

          [edit] General Summary and Thoughts (Mai)

          [edit] Drawing ROIs

          • By far, the most difficult ROI to draw is for the UNC. Originally, we followed Mori's protocol by delineating the ROI in the rgb-vector filter. However, when we looked at the ROI in the t1 image, it became clear that the ROI was missing a significant number of voxels in the lateral inferior border. The rgb vector image is very messy, and it's sometimes difficult to pick out borders, so we switched to drawing on the t1 image, using the same protocol. This didn't work either because when compared to the rgb vector, the new ROI missed large portions of UNC. Though we dropped the UNC for now, in the future it might be worth redefining the ROI to extend all the way to the slyvian fissure, as that would include the entire UNC.
          • The ILF ROI 2 can also be difficult to draw. Firstly, choosing the correct slice can be ambiguous since it is not always clear when the temporal lobe is completely seperate. Referencing the MD image can help, but it too can be ambiguous. Secondly, the temporal lobe is often surrounded by blood vessels, which can be difficult and annoying to avoid. Thirdly, sometimes the inferior border is poorly defined because the image becomes very blurry. Finally, the medial border can be difficult to define.
          • Occasionally, the posterior most point of the POS is difficult to determine.

          [edit] Alignment Issues

          • In general, dti images and anatomical images are fairly well aligned. In most brains, expect to see misalignments in no more than 2-3 different ROIs per hemisphere. Misalignments are generally between 1-3 voxels along one border. More than 4-5 voxels is highly, highly unusual. Misalignments on more than one border are fairly unusual as well.
          • ILF ROI 1 sometimes has ventricles present, especially in adolescents.
          • ILF ROI 2 often has misalignment along the inferior border, with the ROI extending beyond the brain mask.
          • IFOF ROI 2 has the most severe misalignment, across many brains and both hemispheres. Usually, it occurs along the inferior border. The medial most point of the ROI extends off the dti image, while towards the middle of the hemisphere, it is several voxels short. This implies an uneven stretching or misalignment of the images.

          [edit] Classifying via Mori Atlas

          The process delineated above relies on manually setting planes via ROIs and then tracking between them. Preliminary analysis using this method, however, has shown that it can be somewhat unreliable, missing many fibers because they don't extend all the way from one ROI to the other. It is also a highly time-consuming process.

          An alternate process is to automatically classify fibers using the Mori fiber Atlas. This process has been shown to be as reliable, if not more so, than manual tracking. The Mori Atlas was created by comparing several hundred "normal" brains and calculating the probability that each voxel is part of a certain major fasicle. These brains were then normalized onto a single image that combines the probabilistic data from the normal brains: e.g. Voxel A is 50% of the time part of the ILF, 20% F major, etc.

          The lab has code, dtiFindMoriTracts, that will automatically classify the fibers in a subject's brain according to the Atlas.

          1. Based on the tensor shapes and other data, individual fiber tracts will be generated, resulting in a mass of undifferentiated fibers.
          2. Each voxel in a given tract will be compared to the corresponding normalized voxel in the Mori Atlas, extracting the probabilistic tract data.
          3. Once all the voxels in the tract have been compared, all the probabilities will be averaged. Note: In future code, consider modifying code to weight different probabilities or voxels more (e.g. We might want to weight a voxel that is equally likely to be a whole bunch of different major fasicles lower than a voxel that is almost all of the time a single fasicle).
          4. Choose the highest probability to classify the fiber. Note: Currently the code is set to classify ALL fibers (threshold = 0). It may be that a fiber is NOT part of a fasicle, though, so we may want to later change the classifying threshold.
          5. Repeat for all the fibers.

          Relevant codes and scripts for performing analysis:

          • Function for classifying Mori fibers. Minimum input values are setting the base directory and outFile. Includes example script calling dtiFindMoriTracts.m after return: dtiFindMoriTracts.m
          • Another example script calling dtiFindMoriTracts: Z:\Kids\dti\moriTracts\dti_FFA_findMoriTracts.m
          • Function that calculates fiber group properties based on Mori classified fibers: dtiGetFgProperties.m.
          • Example script that calls dti_FFA_getFgProperties.m, a modified version of dtiGetFgProperties.m. There are currently problems with memory space when calling this using this script: Z:\Kids\dti\moriTracts\dti_FFA_getFgPropertiesScript.m

          [edit] Elena notes 03/12/2010

          To be organized on the wiki later

          • Normal procedure is to use first the ROIs and then the template to classify fibers
          • Heidi Johansen-Berg and maybe others have also used a winner-takes-all classification approach, but not based on using probabilistic tracking information along each fiber, but instead by probablistic tractography and endpoint frequencies
          • To check the normalization: make sure the code generates a MoriGroups_snCheck.png image -- the template is the red channel, the individual subject is cyan (green and blue channels) -- we are looking for big errors (top and bottom slice misalignments are OK).

          [edit] References and validation

          DY will check with ER we are reading the correct references for the atlas we are using.

          In-house validation attempts: Elena and Michael did some manual parcellation and came up with similar results to the automatic procedure. However, for the automatic procedure, they did use the ROIs AND the template.

          Jason Yeatman feedback: Jason also compared his own manual arcuate definitions with the automatic arcuates from this procedure. I have emailed him to find out what he concludes from this comparison. His own procedure is kind of "taking the spirit of the Mori planes" but not actually following the exact procedural details. His intuition is that using the Mori plane criteria on an MNI brain (+ some manual editing) would be the best approach.

          My current opinion: The problem I've had with the Mori planes on individual subjects is that the criteria for plane selection can really vary relative to gross anatomical position (e.g., the slice where the temporal lobe is detached from the frontal lobe falls relative to the pole in very different locations in different individuals -- whereas selecting based on MNI coords would hopefully yield similar locations relative to the gross proportions of the brain A/P/L/M). BUT -- I'm generally wary of the idea of only classifying a fiber based on two points. A further question is whether I want to combine this with the probabilistic atlas so that classification is not just based on two points, but along the entire trajectory. I can compare the fiber classification outcomes if I just use the MNI ROIs, versus just the atlas, versus both in some particular order.

          The atlas is: MNI_JHU_tracts_prob.nii.gz in vistasoft/trunk/mrDiffusion/templates -- this is a preloaded atlas that comes with the FSL package, and is further described on the FSL white matter atlas page.

          There is also a symmetrized version in the same location: MNI_JHU_tracts_prob_Symmetric.nii.gz. This was created by flipping the hemispheres, and taking an average of the two.

          Hua et al, 2008 appears to be a preliminary version of the Mori et al, 2008 paper. It describes a probabilistic atlas created from 28 subjects, and includes only 11 fiber groups. This appears to be the primary reference for details on how the MNI_JHU_tracts_prob.nii.gz atlas was created. Zhang et al, 2008 describes a method for placing Wakana-style ROIs on the ICBM-DTI-81 atlas in MNI coordinates, which can then be transformed to individual native space and used to define tracts -- this yields nearly identical results to the results from manual ROI placement.

          Oishi et al, 2009 describes a method for determining parcellation accuracy by comparing it to inter-rater manual reliability and showing that they are equivalent -- in other words, the automatic procedure is just as similar to an expert rater as another expert rater.

          TODO: become a knowledgeable expert on all the different ways major fascicles are defined, both manually, and automatically -- and on the different JHU/UCLA white matter atlases and their potential uses. Search: "dti atlas" or "dti template".

          TODO: Find out how exactly the dtiFindMoriTracts code is using the MNI ROIs and template in combination.

          TODO: Try tracking with the MNI ROIs. Figure out who the VIEW the MNI ROIs. They are in mrDiffusion/templates. They can theoretically be loaded by: File>ROIs>Load Rois from MNI Nifti -- although when I tried it just now they broke. Maybe I want to make my own ROIs, just to see if I can? Or see if they can be downloaded from somewhere (based on Mori refs)?

          [edit] Editing wmProb.nii.gz

          DTI data from brain stem and cerebellum tend to be of low quality and are, in this case, irrelevant areas of study. Before determining the quality of DTI images to ensure comparable images in different age groups, these areas must be removed so that they do not affect quality analysis.

          [edit] Mai to-dos 2010

          • Check that all subjects have all files (except four adolescents: RCV, CC, JPB, KWL) -- notify DY if anything missing
          • Start plotting/comparing data
          • Kids\fmri\code\ROIsize\all_roi_size.m: this code links behavioral measures with fMRI measures; our goal is to write code linking the same behavioral measures with DTI measures, so we want to understand and eventually modify this code for our purposes. Status: Mai has read and annotated a version of this code called, 'all_roi_size_Mai_notes.m'. She also has an email listing the overall sequence of steps.

          Future to-dos

          • Look for hemispheric differences (e.g., LH ILF vs RH ILF)
          • Compare age groups (adults vs adolescents) and/or partial out age
          • Then correlate these measures with behavioral performance and age

          Grant references

          1. Thomas et al, 2008
          2. Lebel et al, 2008
          3. Dougherty
          4. Golarai et al, 2008
          5. Golarai et al, 2007

          [edit] Davie to-dos 2010

          • Make plan to define LH ILFs
          • RESTORE / non-RESTORE comparisons -- all kids, plus adolescents RCV, CC
          • Fix kid TEL and adol JPB
          • Capgras: pick times for Mai to do first and second rounds of segmentation

          [edit] Notes

          • ER: when using group defined MNI ROIs and the dtiFindMoriTracts.m procedure, you have to worry about tracts not quite reaching the group defined plane ROIs, particularly for the ILF and IFOF. This is why her code includes threshold parameter.
          • MP has looked at many brains and has developed a heuristic procedure (that only he knows and is not written down) for defining these tracts in individual brains.
          • Er points out that we should not limit our analysis necessarily to fiber groups only included in Mori's procedure. There are many other fiber groups that can be identified but have low Mori reliability.


          [edit] References

          Based on ER code, mrDiffusion/fiber/dtiFindMoriTracts

          [edit] Cibu Thomas conversation 12/9/2009

          Full notes in lab notebook

          • Smoothing: removed need for manual cleaning on laborious way (still cleaned forceps); procedure described in Westin 2007, in an executable patented by author Kwanjin Jung
          • Tracking: Seeded/tracked all voxels FA>0.2, angle thresh 40 degrees (higher, e.g., 60, would result in too much manual cleaning
          • ROI pairs: independently developed by Cibu, not baed on Mori, but largely similar
          • Fiber percentages: Number of fibers divided by number of all fibers in whole brain > 8 pixels (8 pixels x 3mm voxels)
Personal tools