fidoptPancreatic-CT-CBCT-SEG
Breath-hold planning CT (fixed) and cone-beam CT (moving) of one patient treated by radiotherapy for pancreatic cancer (Pancreatic-CT-CBCT-SEG).
Read before comparing (5)
- Alignment. The provider aligned the CBCT rigidly on an implanted fiducial marker or a surgical stent, then resampled it onto the planning CT's grid; the pair is used as delivered. That marker, visible on both images, is a possible point truth not used yet: its starting error would be close to zero.
- Truth. On the CBCT only the parts of the organs inside the volume of interest (the low-dose PTV plus 1 cm) were contoured, on the planning CT up to 2 cm past it. Both maps are cut to the planning CT's structure "ROI", taken to be that volume, and to the slices the CBCT covers, which also evens out the CBCT contours that run past it in the files.
- Scope. The board therefore scores the stomach and bowel near the target only: 12 to 141 cm³ a label on the test pairs.
- Images. The planning CT has iodinated contrast, the CBCT none. The CBCT covers about 18 cm along the body and 25 cm across, though four of the ten test CBCTs are about 45 cm across; outside its field it holds -1000. Its values vary more: inside the labels its medians on the test pairs run from -166 to 149, the CT's from 28 to 83.
- Not used. The second CBCT of each patient, the original CBCTs, the repeat contours of two observers on ten patients, the RT doses and the lung contours.
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 | 0.671 ±0.075 | 12.12 ±3.51 | 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
- Test pairs
- 10 test patients drawn with seed 2026 among the 30 outside development, and 10 development patients. Each gives the first of its two CBCTs by date, so no planning CT is scored twice.
- Truth
- contours by experts the stomach with the first two segments of the duodenum (1) and the rest of the small bowel (2), contoured on both images by radiation oncologists and reviewed by two medical physicists, cut to the volume of interest
- Source
- www.cancerimagingarchive.net/collection/pancreatic-ct-cbct-seg
doi.org/10.7937/TCIA.ESHQ-4D90
doi.org/10.1002/mp.14906 - Licence
- CC BY 4.0
10 pairs for tuning, never tested
Pancreas-CT-CB_001, Pancreas-CT-CB_002, Pancreas-CT-CB_003, Pancreas-CT-CB_030, Pancreas-CT-CB_008, Pancreas-CT-CB_038, Pancreas-CT-CB_007, Pancreas-CT-CB_023, Pancreas-CT-CB_010, Pancreas-CT-CB_018