fidoptLV infarct scar on LGE MR (MICCAI 2012 challenge)
A short-axis late gadolinium-enhanced MR stack of the left ventricle of a patient or a pig after a myocardial infarction, the myocardium given as a mask and the scar outlined by a consensus of three cardiologists (Karim et al. 2016).
10 test casesheartimage MR
Read before comparing (8)
- Scans. Five patients with ischaemic cardiomyopathy at least three months after an infarction, on a 1.5 T Philips Achieva at King's College London, and five pigs six weeks after occlusion of a coronary artery, on a 3 T Siemens scanner at KU Leuven (Karim et al. 2016). Short-axis LGE stacks as delivered: 0.63 to 1.37 mm in plane, 2 to 10 mm between slices for the patients, 1.72 x 1.72 x 6 mm for the pigs, 16-bit values without unit; acquired at 1.8 x 1.8 x 8 mm (patients) and 1.8 x 1.8 x 6 mm (pigs); the patients' stacks are delivered on finer grids. Scores average five patients and five pigs; the challenge also reported them apart.
- Headers. The patients' files and pig a1's carry an identity direction (origin zero for the patients), so their orientation in the body is unknown; pigs a2 to a5 carry the scanner's oblique direction. The voxels are untouched.
- Mask. The myocardium mask, drawn by an expert between the epicardial and endocardial contours without the papillary muscles, is given to methods, as the challenge gave it: the task is the scar within it.
- Truth. Three cardiologists outlined the enhanced myocardium on each slice over that mask, blinded to the clinical history; STAPLE merged them and the consensus is its probability above 0.7. Graded expert. The individual observers' outlines are not released.
- Scar. 4.6 to 24.5 mL on the patients, 5.2 to 12.4 mL on the pigs; 1 to 11% of a pig's drawn scar lies just outside its myocardium mask (none of a patient's), left out of the truth, as the challenge scored only inside the mask.
- Start. Inside the myocardium mask, the whole mask scores a mean Dice of 0.17 against the scar, Otsu's bright part 0.49 and the full width at half maximum (half the brightest myocardial value) 0.50; 0.07 to 0.74 a case. The truth moved one voxel in plane keeps 0.69 to 0.93 (0.79 on average).
- Overlap. No other descriptor names this record or its challenge. LASC2013 also holds King's College London scans on a Philips 1.5 T, but whole-heart bSSFP of the left atrium in atrial fibrillation; neither release keys its patients, so a shared patient cannot be excluded, only no shared image.
- Leakage. The ten scans and their scar references have been public as the challenge's training set since 2012, on figshare since 2017: a model trained on them has seen every test case.
No method has a result on every case of this dataset yet.
Case by case
Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- All ten scans with a released scar reference are test cases, the challenge's training set of five patients and five pigs: none is left for development. The challenge's twenty test scans have no released scar reference.
- Truth
- contours by experts the scar consensus of the challenge: three cardiologists outlined the enhanced myocardium slice by slice over the given myocardium mask, merged by STAPLE and thresholded at a probability of 0.7
- Source
- doi.org/10.6084/m9.figshare.4214622.v1
www.doc.ic.ac.uk/~rkarim/la_lv_framework/lv_infarct/index.html
doi.org/10.1016/j.media.2016.01.004 - Licence
- CC BY 4.0