fidoptCardiac cine MR of pigs at 7 T (cmr-cine-sscrofa)
Short-axis cine MR of one pig's heart at 7 T, end diastole (fixed) and end systole (moving), 12 days after an infarction or a sham procedure, the left ventricular cavity and myocardium drawn at both.
10 test pairsheartsame patientfixed MR, moving MRdelivered aligned
Read before comparing (9)
- Cohort. Eleven pigs, seven given an infarction by a 90-minute balloon occlusion of the left anterior descending artery and four a sham procedure, each scanned before and 3 ± 1, 12 ± 1 and 58 ± 1 days after.
- Truth. Observer one's first reading, the paper's reference: the endocardial and epicardial borders drawn by hand in Medis QMass, the papillary muscles left in the cavity. On the ten test cases observer two agrees at a Dice of 0.89 for the cavity and 0.84 for the myocardium, observer one's second reading at 0.94 and 0.90: no method should be expected far past that.
- Coverage. Only the left ventricle is labelled. One or two basal slices are unlabelled on 8 of the 10 test cases and one or two apical slices on 4; both frames are labelled on the same slices. On some apical slices the observer drew the epicardium alone, so the whole section is myocardium (11 slices and frames on 5 test cases).
- Slices. 12 to 14 slices of 0.425 mm pixels, 6 mm thick and 6 mm apart, each a 2D drawing: the labels check motion in the plane better than across it, where a field gains or loses a label a whole slice at a time. The drawn myocardium changes volume between the frames, by +2% (A38) to +27% (A09).
- Acquisition. Retrospectively gated 2D gradient echo, 30 frames a cycle, under breath hold, each slice its own acquisition: neighbouring slices can be offset by different breath-holds, the same offset in both frames. The two frames of a slice share one acquisition and grid, used as delivered.
- Artefacts. 7 T adds susceptibility at the posterior wall, signal voids, and gating disturbed by the arrhythmias that follow an infarction.
- Masks. The provider's, drawn from the Medis contour points: half a pixel (0.2 mm) of doubt in the plane. Slice z of a mask is the series the provider's naming gives it: on the 11 scans built the blood pool under the cavity label is 1.1 to 2.6 times brighter than the wall, and reversing the slice order on A10 brings it down to 0.6.
- Baseline. Between the frames the cavity loses 34% to 42% of its volume on the infarcted pigs and 54% to 63% on the sham ones. With no registration the labels overlap at a mean Dice of 0.61 (0.46 to 0.71), 0.67 on the infarcted pigs and 0.52 on the sham ones, and an HD95 of 9.6 mm (8.0 to 12.0).
- Outside numbers. None: the paper is about segmentation, and the pairing as a registration is ours.
No method has a result on every pair of this dataset yet.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| No registration | 0.609 ±0.062 | 9.55 ±0.90 | 0.000 | 0.00 | ||||||
Dice: how much the organ contours overlap, 1 is perfect. HD95: distance between the contours' surfaces. 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
The first test caseClick to look at every case.
- Test pairs
- One scan a pig, its third, 12 days after the procedure, the latest all eleven pigs have; the 29 other scans of these pigs stay out. 10 test pigs drawn with seed 2026 by group, six infarcted and the four sham ones; the seventh infarcted pig, A26, is the development case.
- Truth
- contours by experts the left ventricular cavity (1, the papillary muscles inside) and the myocardium (2), within the endocardial and epicardial contours observer one drew by hand on the short-axis slices of the two frames (the provider's masks of that reading)
- Source
- zenodo.org/records/7684034
github.com/chfc-cmi/cmr-cine-sscrofa
doi.org/10.1038/s41598-024-61417-4 - Licence
- CC BY 4.0 (the repository asks that the corresponding publication be cited: Kollmann et al., Scientific Reports 2024)
1 pair for tuning, never tested
A26