fidoptPDDCA
Head and neck CT of two patients treated by radiotherapy for stage III or IV cancer of the oropharynx, hypopharynx or larynx, paired by us from PDDCA 1.4.1 (RTOG 0522 trial; MICCAI 2015 challenge).
Read before comparing (7)
- Truth. The PDDCA authors redrew the contours on current RTOG protocols and published guidelines, not from the trial's structure sets. The chiasm is hard to see on CT and was drawn from anatomy; the brainstem is cut on one axial slice at each end, with about one slice of doubt at the bottom.
- Landmarks. Five bony points on the mandible, the dens and the occipital bone, placed for gross alignment: they say nothing of the soft tissue and give a coarse error.
- Missing contours. Some patients lack the mandible, others one or both submandibular glands, upstream. A pair is scored on the organs both patients hold.
- Alignment. Each scan keeps its native grid, the world frames up to 1.4 m apart. The moving header carries one translation from elastix on the intensities; no contour or landmark sets the start.
- Baseline. With no registration the ten test pairs score a mean landmark error of 19.9 mm (11.1 to 39.3) and a mean Dice of 0.201 (0.011 to 0.422). 0522c0125_0522c0057 (39.3 mm) and 0522c0455_0522c0079 (36.9 mm) start furthest.
- Field of view. The fixed scan of 0522c0125_0522c0057 ends one to three slices above its chiasm, so the other patient's chiasm and optic nerves fall outside it and have zero overlap; all nine organs stay in the Dice. The scans come from many RTOG 0522 sites: a pair may join two scanners and two lengths of neck.
- Outside numbers. None: the pairing is ours.
| Accuracy | Smoothness | Cost | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | TRE mm | rTRE | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 1 | ConvexAdam (MIND)GPUapp | 0.519 ±0.089 | +0.318 | 10.66 ±5.70 | 7.92 ±3.76 | 0.0104 ±0.0058 | 0.141 | 1.83 | 34 | 4.3 | 9.9 | |
| 2 | DemonsCPUapp | 0.344 ±0.117 | +0.143 | 13.69 ±6.45 | 13.50 ±6.94 | 0.0180 ±0.0103 | 0.120 | 1.40 | 31 | 5.2 | ||
| No registration | 0.201 ±0.086 | 19.05 ±7.10 | 19.90 ±7.51 | 0.0263 ±0.0114 | 0.000 | 0.00 | ||||||
The order averages each method's ranks on Dice, HD95 mm, TRE mm, SDlogJ, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 10 pairs. Two neighbours are tied when a paired Wilcoxon test on the shared pairs does not find the upper one better on Dice (5%): the places a method could hold are then shown under its own. The best value of each column is in colour. Click a row to see its images, a column title to sort by it. Time and memory measured on Laptop.
Dice: how much the organ contours overlap, 1 is perfect. HD95: distance between the contours' surfaces. 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 pairs drawn with seed 2026 among the 10 outside development, and 6 development pairs. Only 33 of the 48 patients have landmarks, the challenge's training and additional training cases; they are paired, each in one pair only, and one is left over.
- Truth
- contours by experts landmarks by experts the dataset's manual contours of nine organs at risk: 1 brainstem, 2 optic chiasm, 3 mandible, 4 left and 5 right optic nerve, 6 left and 7 right parotid, 8 left and 9 right submandibular gland the dataset's five manual bony landmarks: chin, right and left mandibular condyle, odontoid process, occipital bone
- Source
- www.imagenglab.com/newsite/pddca
www.imagenglab.com/data/pddca/pddca.odt
doi.org/10.1002/mp.12197
doi.org/10.7937/K9/TCIA.2015.7AKGJUPZ - Licence
- none stated: PDDCA calls itself 'A Public Domain Database for Computational Anatomy' and says its CT were 'made available to the public domain' through TCIA (Head-Neck Cetuximab)
6 pairs for tuning, never tested
0522c0002_0522c0190, 0522c0009_0522c0003, 0522c0014_0522c0329, 0522c0457_0522c0328, 0522c0077_0522c0441, 0522c0253_0522c0330