fidoptSTACOM 2011 cardiac motion
3D tagged MR of the heart of a healthy volunteer at end diastole (fixed) and end systole (moving) of one cycle (STACOM 2011 cardiac motion analysis challenge).
Read before comparing (6)
- Truth. Each of the 12 landmarks sits on a tag-line crossing in one wall (anterior, lateral, posterior, septal) at one level (basal, mid, apical) of the left ventricle, tracked by hand through the cycle by two observers. The truth is their mean, or observer 1 alone where the organisers excluded observer 2's track (5 tracks of the test volunteers: v11, v12, v13).
- Observers. At end diastole the two start from the same points on 7 of the 10 test volunteers, so that frame says nothing of their agreement. At end systole they are 0.97 mm apart on average (median 0.87 mm, at most 7.0 mm on one landmark of v15), for 8.7 mm of motion at the landmarks.
- Coverage. 12 points a pair, all in the left ventricular wall: the board says nothing of the right ventricle, the blood pool or the tissue around the heart.
- Alignment. The two frames come from one acquisition on one grid, so nothing is aligned. Volumes and landmarks are in their own frame, a VTK origin and spacing without orientation: the axes follow the acquisition, not the patient's LPS axes.
- Images. Three tagged acquisitions, one for each orthogonal direction, taken in three breath-holds and fused: 3 T, tags 7 mm apart, 0.96 mm voxels around the left ventricle. At end diastole the blood is magnetised like the tissue and shows tags, at end systole it shows a blood pool, so intensities change between the images. The end-systolic frame is the challenge's own.
- Outside numbers. The challenge's four methods reached a median error of 1.1 to 1.5 mm on the volunteers, but over all frames of a tracked sequence, not on one pair.
No method has a result on every pair of this dataset yet.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | TRE mm | Change | rTRE | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| No registration | 8.70 ±0.62 | 0.0466 ±0.0033 | 0.000 | 0.00 | ||||||
TRE: distance between matching landmarks. rTRE: the TRE over the length of the image's diagonal, to compare datasets of other sizes and units. SDlogJ: how irregular the deformation is, 0 is perfectly smooth. Folds: share of the image where the deformation folds over itself.
Pair by pair
Click a score to look at that result: the fixed image, and over it the moving image after registration, with the annotations. Underlined scores have their image; for the others you see the pair without the result. A green cell gained against no registration on that pair, a red one lost.
About the data
- Test pairs
- 10 test volunteers drawn with seed 2026 among the 12 outside the first 3 development volunteers; the 2 left over, v7 and v10, joined development, which holds 5. The phantom (8 landmarks) is left out.
- Truth
- landmarks by experts the dataset's 12 landmarks of the left ventricular wall, tracked by hand by two observers; the mean of the two
- Source
- www.cardiacatlas.org/motion-tracking-2011-challenge
www.cardiacatlas.org/wp-content/uploads/2024/03/readme.pdf
doi.org/10.1016/j.media.2013.03.008
inria.hal.science/hal-00855928 - Licence
- open access by attribution (cite Tobon-Gomez et al. 2013)
5 pairs for tuning, never tested
v1, v2, v4, v7, v10