[wiki:FsTutorial top] | [wiki:FsTutorial/MorphAndRecon previous]

Troubleshooting Guide

This guide provides troubleshooting information for common problems in cortical reconstruction.

1.0 Problems creating surfaces

1.1 Skull stripping

Refer to the [wiki:FsTutorial/SkullStripTutorial Skull Stripping exercise] for troubleshooting guidelines.

1.2 Failure of automatic cutting planes

In some cases, the cutting planes are not found correctly during the white matter fill step of the volume processing pipeline. The results of this can be seen in [wiki:FsTutorial/Tools tkmedit] as incorrectly colored voxels in the filled volume, as straight lines making inappropriate cuts in the volume, or in the inflated surface using tksurfer. Automatic cutting plane failure can also be seen in the wm volume, where the brain stem is being incorrectly identified as white matter, and in tksurfer where there is a large protrusion in the center of the medial aspect of the inflated surface.

The following instructions describe how to manually define the two cutting planes to correct these problems.

1.2.1 Left-right hemisphere separation (corpus callosum cutting plane)

Load the subjects WM volume into [wiki:FsTutorial/Tools tkmedit] with the T1 volume as the aux volume, and look for a sagittal slice in the wm volume in which the corpus callosum is disconnected from the rest of the wm. This slice should be near the mid-line. If you cannot find a slice with the corpus callosum already disconnected, then detach it manually by erasing a fine line of voxels around its perimeter. Select Edit Voxels under [wiki:FsTutorial/Tools tkmedit]’s Tools menu, then click with button-3 to erase (see ‘Guide to [wiki:FsTutorial/Tools tkmedit]’ for more information on editing voxels). Then select Save Main Volume under the File menu.

Now pick a point (mouse button-1) in the corpus callosum and verify with both the horizontal view and the coronal view that your cursor is central in all three views – adjust as necessary. Once you have a good point you can view the MNI coordinates by going to View -> Information -> MNI coordinates. If these are not available use the Volume Scanner coordinates, viewed in the same way. These coordinates, available in the Cursor section of the [wiki:FsTutorial/Tools tkmedit] Tools window, can be used with the following command to run the recon-all process starting from the mri_fill step:

recon-all -autorecon2-wm -cc-xyz <X Y Z> -subjid <subject name>

If the pons cutting plane also needs to be set, see below for instructions and then you can use one command, with two sets of coordinates, to run recon-all from the mri_fill step.

1.1.2 Separation of brain stem and cerebellum (pons cutting plane)

Load the subjects WM volume into [wiki:FsTutorial/Tools tkmedit] with the T1 volume as the aux volume and find a horizontal slice in the WM volume in which the brainstem is disconnected from the rest of the brain. This slice should be near the top of the pons. If you cannot find a slice with the mid-brain disconnected, then separate the mid-brain manually by erasing a thin line of voxels by clicking button-3 in [wiki:FsTutorial/Tools tkmedit], then save your changes to the WM volume by choosing File -> Save Main Volume. Click on a point (mouse button-1) and verify with both the horizontal view and the coronal view that your cursor is central in all three views – adjust as necessary. Once you have a good point you can view the MNI coordinates by going to View -> Information -> MNI coordinates. If these are not available use the Volume Scanner coordinates, viewed in the same way. These coordinates, available in the Cursor section of the [wiki:FsTutorial/Tools tkmedit] Tools window, can be used with the following command to run the recon-all process starting from the mri_fill step:

recon-all -autorecon2-wm -pons-xyz <X Y Z> -subjid <subject name>

If both the corpus callosum and pons cutting planes need to be set you can use one single command to do this:

recon-all -autorecon2-wm -cc-xyz <X Y Z> -pons-xyz <X Y Z> -subjid <subject name>

1.2.3 Cerebellar attachment

Sometimes after mri_fill is run, creating the WM volume and inflated surface, you will find that the cerebellum is attached to the rest of the cortex. This is usually due to failure of the automatic cutting planes, as shown in the previous section, but sometimes the segmented hippocampal strands are joined to the cerebellum. These adjoining voxels must be manually erased. Load the WM volume into tkmedit and select Tools -> Edit Voxels and using mouse button-3 you can remove the necessary voxels. After saving your changes, rerun the surface creation steps using the command:

recon-all -autorecon2-wm -subjid <subject name>

1.3 Manually editing topological defects

Refer to the [wiki:FsTutorial/FixingDefects Manual Edits exercise] for troubleshooting guidelines.

1.4 Automatically fixing surface topology

On occasion, you may find that the topology fixer seems to fail, snarling the surface in regions quite a distance from the area of your edits. The reason for this is that the topology correction procedure assumes that a defect is small relative to the "correct" topology of the surface. If this is not the case, either because there is one large defect or a number of smaller ones close to one another, then the topology correction may result in a local geometric inaccuracy in the surface. Note that the topology of the surface is guaranteed. You can get the mris_fix_topology binary to write out the labels (i.e. the sequentially numbered contiguous regions of what it detected as defects) on the original surface by typing:

setenv DIAG 0x04048
setenv DIAG_VERBOSE 1
mris_fix_topology -suffix _fixed <subject name> <hemisphere>

This will cause the topologically correct surfaces to have the suffix "fixed" appended to their names (e.g. lh.orig_fixed), so that the unfixed surfaces will not be overwritten. You should be able to bring up the unfixed surface in tksurfer (e.g. tksurfer <subject name> <hemisphere> inflated) and read in the curvature file named lh.defect_labels or rh.defect_labels. This will show you exactly what the topology correction detected as defects.

Buckner40 Inspection Notes

The 40 subjects in the [wiki:FsTutorial/Data buckner_data/group_study tutorial distribution] have each been inspected for defects by JenniPacheco and EvelinaBusa. These notes, combined with using tkmedit and tksurfer to observe the defects, serve as a good tutorial on what are considered 'good' and 'bad' scans.

Note: Some of these subjects are used as the example cases in the Morphometry and Reconstruction tutorial (specifically, subjects 008, 039, 095, 108, and 111)

No.

Inspection notes

004

LH temp lobe surfaces aren't good (coronal slice 116)

008

LH white surf in lateral vent not good (coronal slice 67)

RH pial surf misses a lot of temp lobe (slice 147)

017

LH surfs in lateral vent not good (coronal slice 49)

021

LH artery is little island of pial surface (coronal slice 139)

032

not bad overall (pial in frontal lobes could extend a bit)

039

bad talairach alignment

040

LH pial surf a bit tangled (slice 92 - vol index 145 142 92)

045

superior CC cut plane too much in LH

RH pial sur misses a lot of temp lobe (slice 139)

LH pial sur misses a lot of frontal & temp lobe (slice 146)

049

not bad overall

067

good overall

073

good overall

074

temp lobes - lose some pial and white surfs (slice 147)

080

very bad, most of inferior brain surfs are missing

084

good overall

091

not bad overall

092

good overall

093

good overall

095

very bad - CC plane in medial LH, pial surf in RH wonky (slice 68)

097

RH temp lobe loses pial (slice 146), LH pial wonky laterally (166)

099

RH frontal lobe surfs on slice 150? (132 165 150) otherwise good

102

good overall

103

a chunk of RH artery in pial surface (slice 145) otherwise good

106

good overall

108

very bad -- RH cereb missing, RH temp lobes losing surfs (145)

inferior RH occip lobe labeled as cereb cortex in aseg 95 146 87

111

temporal lobe: disconnection of thin gyri, slices 133-135

frontal lobe: misclassification of non-wm as wm, slice 179+-10

lateral ventricle: not filled, slice 50-168

114

not very good - overall loss of outer pial surface

123

not bad overall

124

good overall

128

not bad overall

129

not bad overall

130

good overall

131

good overall

133

not bad overall

136

good overall

138

not bad overall

140

good overall

141

good overall

144

very bad - CC cutting plane in medial RH, LH lat ventr surfs iffy,

artifacts in scan (e.g. slice 125)

145

bad seg - orbital fat labeled as RH wm (slice 173), L/RH temp lobe, surfs 143

149

not bad overall