This is the summary of the LV segmentation challenge results that has been presented at the STACOM 2011 workshop. More details of this result can be seen in [Suinesiaputra et. al., STACOM 2011].
We received initial 38 requests for participation from 29 different research institutions. Only 5 participants submitted their results and 3 of them published their papers at the STACOM 2011 workshop. Here are the characteristics five participants, or the auto raters:
|Rater ||Description ||Type |
|SCR||Deformable registration method||2D feature recognition|
|INR||Voxelwise classification technique||3D models|
|AO||Optical flow algorithm||2D lines/pixels|
|DS||Contour tracking algorithm||2D lines/pixels|
|EM||ACM + optical flow method||3D models|
There were 2 expert raters: AU & NU raters. Their characteristics are shown below.
| ||AU Rater ||NU Rater |
|Type||3D models||2D contours|
|Interaction||Supervised FEM fitting||Manual delineation|
|Timeframe||Full cardiac cycle||ED & ES|
|View||Any slice orientation||Short-axis view|
|Clinical values||Global LV functions||Not available|
|Source||University of Auckland, NZ||Northwestern Univ., USA|
With this expert rater characteristics, we defined the following validations:
- Validations on clinical values and performances for the participant submissions of a subset (or full set) of the validation set were made by using the AU rater as the reference.
- Validations on clinical values for the collation study were made by using the AU rater.
- The ground truth images estimated by using the STAPLE method took all raters (2 experts and 5 auto) as the input from the overlapped 18 cases and from ED & ES of the short-axis view only.
|Rater ||EDV diffs (ml) ||ESV diffs (ml) ||Mass diffs (gr) |
|SCR||13.03 (± 18.13)||18.98 (± 16.20)||-9.56 (± 22.58)|
|INR||-79.88 (± 39.62)||-61.96 (± 38.83)||74.75 (± 53.74)|
|AO||8.69 (± 99.39)||13.36 (± 64.16)||51.49 (± 95.65)|
|DS||3.94 (± 23.14)||25.65 (± 17.71)||1.08 (± 28.40)|
|EM||-77.75 (± 50.60)||-45.46 (± 36.44)||-51.92 (± 39.92)|
Region Operating Characteristics (ROC) curves