AutomSegPipe:ComparisonStudy

FLASH/MPRAGE tissue segmentation reliability study on BullittStudy data

This dataset is composed of 97 cases, classified by age group (20-29, 30-39, 40-49, 50-59, 60+ years old) and gender. 84 cases have been computed to get an automatic segmentation of probabilistic subcortical structures, the related pipeline using a dual-channel tissue segmentation.

This page describes a dual channel tissue segmentation reliability study. Two seperate trials have been computed, which are a T1_FLASH&T2 vs a T1_MPRAGE&T2 tissue segmentation. Identic parameters have been used for both trials:

• Tissue Segmentation parameters:

- itkEMS: 1.5
- Filter Iterations: 10  - Filter Time Step: 0.01  - Filter Method: Grad aniso diffusion  - Max Bias Degree: 4
- Prior 1(WM): 1.3       - Prior2 (GM): 1.0        - Prior3 (CSF): 0.7                    - Prior4 (other): 1.0

Tissue segmentation based on T1_FLASH and T2 images

84 cases have been computed, whose source images are both T1_FLASH and T2 images.

• Script location: /BullittStudy/Processing_AutomaticSegmentation/Processing_T1FLASH/AutoSeg.script
• Quality Control location: /BullittStudy/Processing_AutomaticSegmentation/Processing_T1FLASH/QCPic-{AutoSeg,EMS}
• Data location: /BullittStudy/AgeGroup*/*/NormalA-*/AutoSeg/...

Tissue Segmentation based on T1_MPRAGE and T2 images

10 random cases have been selected (one for each age group and gender) to compute this dual-channel segmentation, based on T1_MPRAGE and T2 images. The related source images are T1_FLASH, T1_MPRAGE and T2 images, in order to compare this tissue segmentation with the previous one.

• Script Location: /BullittStudy/Processing_AutomaticSegmentation/Processing_T1MPRAGE/AutoSeg.script
• Quality Control location: /BullittStudy/Processing_AutomaticSegmentation/Processing_T1MPRAGE/QCPic-{AutoSeg,EMS}
• Data location: /BullittStudy/AgeGroup*/*/NormalA-*/AutoSeg_MPRAGE/...

Methods' comparison

A tissue segmentation volume overlap analysis has been done in order to compare previous methods. Considering each case, Dice and Tanimoto coefficients have been computed for each label: white matter, grey matter and CSF.

• Result Location: /BullittStudy/Processing_AutomaticSegmentation/Comparison_T1FLASH_T1MPRAGE/EMSVolumeOverlap.xls

See results detailed on spreadsheet: Media:EMSVolumeOverlap.xls

The dice coefficients computed for each label are quite low, the average values being:

- WM: 0.56  - GM: 0.5  - CSF: 0.22

Thus only the subcortical structure automatic segmentation using the dual channel T1_FLASH & T2 tissue segmentation has been used so far.