Difference between revisions of "DBP2:UNC:Tissue Segmentation Evaluation"
| Line 14: | Line 14: | ||
A UNC dataset has been used for this study, one patient being scanned 10 times on the same scanner. | A UNC dataset has been used for this study, one patient being scanned 10 times on the same scanner. | ||
Two dual-channel segmentation trials were performed to compare the segmentation reliability using the two algorithms. | Two dual-channel segmentation trials were performed to compare the segmentation reliability using the two algorithms. | ||
| − | A single T1 image was registered to the UNC template atlas and the rest of the T1 images were affinely registered and scaled to that original T1 image to minimize scanner-influenced differences. | + | A single T1-weighted image was registered to the UNC template atlas and the rest of the T1 images were affinely registered and scaled to that original T1 image to minimize scanner-influenced differences. |
=== Results === | === Results === | ||
| Line 27: | Line 27: | ||
See results detailed on spreadsheet: | See results detailed on spreadsheet: | ||
| − | [[Image: | + | |
| + | [[Image:UNCDBP2_TissueSegmentationReliabilityStudy.xls]] | ||
| + | |||
| + | |||
| + | For each algorithm and each tissue, an average T1-weighted label image has been computed. Using the 'EvaluateSegmentationResult' tool, volume differences and tanimoto errors between each case and the computed average have also been computed. | ||
| + | |||
| + | See data detailed on spreadsheet: | ||
| + | |||
| + | [[Image:UNCDBP2_TissueSegmentationReliabilityStudy2.xls]] | ||
=== Screenshots === | === Screenshots === | ||
| Line 34: | Line 42: | ||
== Qualitative study on pediatric dataset == | == Qualitative study on pediatric dataset == | ||
| + | |||
| + | <center> | ||
| + | {| | ||
| + | |[[Image:TissueSegmentation_Pediatric_EMSegment.jpg|thumb|300px|EMSegment tissue segmentation on pediatric case]] | ||
| + | |valign="top"|[[Image:TissueSegmentation_Pediatric_itkEMS.jpg|thumb|300px|itkEMS tissue segmentation on pediatric case]] | ||
| + | |} | ||
| + | </center> | ||
Revision as of 21:49, 12 March 2008
Home < DBP2:UNC:Tissue Segmentation EvaluationBack to NA-MIC Internal Collaborations, UNC DBP 2
Objective
We would like to evaluate the brain tissue segmentation using two different algorithms: itkEMS and Slicer3 EMSegment.
- A reliability study was performed on adult dataset.
- A qualitative study was performed on pediatric dataset.
Reliability study on adult dataset
Method
A UNC dataset has been used for this study, one patient being scanned 10 times on the same scanner. Two dual-channel segmentation trials were performed to compare the segmentation reliability using the two algorithms. A single T1-weighted image was registered to the UNC template atlas and the rest of the T1 images were affinely registered and scaled to that original T1 image to minimize scanner-influenced differences.
Results
One case failed at the execution using EMSegment and has been removed.
The covariance was computed for each trial, not only considering each tissue (White matter, grey matter and CSF), but also the Intra Cranial Volume, sum of the three tissues.
- Covariance/Tissue WM GM CSF ICV - itkEMS 1.27% 0.99% 6.27% 0.88% - EMSegment 7.66% 3.15% 14.84% 1.51%
See results detailed on spreadsheet:
File:UNCDBP2 TissueSegmentationReliabilityStudy.xls
For each algorithm and each tissue, an average T1-weighted label image has been computed. Using the 'EvaluateSegmentationResult' tool, volume differences and tanimoto errors between each case and the computed average have also been computed.
See data detailed on spreadsheet:
File:UNCDBP2 TissueSegmentationReliabilityStudy2.xls
