fidoptPelvic-Reference-Data, CBCT to CT
Treatment cone-beam CT (source) and planning CT (target) of the pelvis of one radiotherapy patient, 4 to 40 days apart, the CBCT registered onto the CT by the corpus (Pelvic-Reference-Data).
10 test casespelvissource CBCT, target CT
Read before comparing (8)
- Pool. 58 patients treated in the pelvic region at Beaumont: a planning CT on a Philips Brilliance Big Bore, then a CBCT for image guidance on an Elekta Synergy 4 to 40 days later (median 14). The CBCT is 88 slices of 3 mm, a 41 cm cylinder reading -1024 outside. Ten of the twenty patients are women, five of the ten test patients.
- Registration. The CT is never resampled: it is cut to the box the CBCT's field of view reaches, and the CBCT is resampled into it once, linearly, through a stored transform. The start is the rigid fit on the experts' bony landmark pairs (1.1 to 2.6 mm at its median), then IMPACT-Reg 1.8.7's Elastix_IMPACT_Jacobian runs with its defaults: a rigid stage and a 10 mm B-spline.
- Exceptions. Patient 028's landmark sheet belongs to other images, so its start is the provider's isocentre shift, from which the registration moved the bones 25 mm. Patient 002's CT points sit one slice off; its start uses them, and the bones moved 6.7 mm. PelvicReference leaves both out; their images are whole, so both are kept here.
- Check. TotalSegmentator on both images: over the structures both hold, the median mean Dice goes from 0.57 after the start to 0.67 after registration, the bones' from 0.69 to 0.82, the bladder's from 0.65 to 0.76, and no kept case lost. Except on 002 and 028 the bones lie 1.3 to 3.3 mm from the landmark fit. Overlays were read where a CBCT is too dark for the network.
- Failures. The rigid stage wanders on the CBCT's weak contrast: on 016, 042, 045, 051 and 052 it moved the box's middle 10 to 23 mm, and the B-spline brought the bones back. On 036 and 048 it went 38.6 and 52.5 mm off and the mean Dice fell to 0.25 and 0.04.
- Residual. The bladder and rectum fill differently on the two days, the bowel gas moves, and several planning CTs hold oral contrast the CBCT does not (002, 012, 016 and 045 among the test cases). The B-spline brings the organs closer without matching them: that error counts for every method alike, and no method reaches an MAE of 0.
- Mask. The CT's body inside the CBCT's field of view as the transform carries it, the couches outside. On large patients the 41 cm field cuts the body at the sides (012, 045), and the mask with it. The CBCT's end slices carry cone-beam artefacts, scored like the rest.
- Baseline. The CBCT's numbers are not calibrated Hounsfield units: inside the mask it reads 119 to 332 below the CT, so the CBCT itself is 235 HU from its CT on average (144 to 348), while a constant at the CT's median inside the mask is 90 HU off (76 to 103). A method must at least take the offset out.
No method has a result on every case of this dataset yet.
Under a licence that is not open
Methods whose code or weights are released for non-commercial use, or under share-alike or custom terms: shown for reference, ranked nowhere.
| Accuracy | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| # | Method | MAE | PSNR dB | SSIM | Time s | RAM GB | GPU GB | Machine | Images |
| KonfAI synthesis app · TotalSynth CBCT to CT, KonfAI appGPUapptrained on CC-BY-NC-4.0 dataolder run | 37.7 ±3.1 | 32.85 ±0.91 | 0.860 ±0.017 | 47 | 3.7 | 4.4 | Workstation | ||
| nnsyn · CBCT to CT, abdomenGPUapptrained on CC-BY-NC-4.0 dataolder run | 45.4 ±6.2 | 31.48 ±1.65 | 0.839 ±0.019 | 3.9 | 2.7 | Laptop | |||
MAE: mean absolute error of a synthetic image against its target, inside the body, in the image's units. PSNR: the same by signal-to-noise ratio, higher is better. SSIM: the same by structural similarity, 1 is perfect.
Case by case
| MAE | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
|---|
Click a score to look at that result: the target image, and over it the method's synthetic image. Underlined scores have their image; for the others you see the case without the result.
About the data
The first test caseClick to look at every case.
- Test cases
- 10 test patients drawn with seed 2026 among the 38 that PelvicReference does not hold, then 10 development patients. 036 and 048, whose registrations failed, were replaced by the next of the seeded order. The collection names its cancers for the whole set, not patient by patient, so the draw has no strata.
- Truth
- target image by a program the patient's planning CT in Hounsfield units, cut to the box of its grid the CBCT's field of view reaches; the CBCT registered onto it by the corpus with IMPACT-Reg 1.8.7 (Elastix_IMPACT_Jacobian, from the rigid fit on the dataset's landmark pairs)
- Source
- www.cancerimagingarchive.net/collection/pelvic-reference-data
doi.org/10.7937/TCIA.2019.WOSKQ5OO
doi.org/10.31579/2640-1053/076
doi.org/10.1002/acm2.12965 - Licence
- CC BY 3.0
10 cases for tuning, never tested
Pelvic-Ref-028, Pelvic-Ref-051, Pelvic-Ref-040, Pelvic-Ref-043, Pelvic-Ref-020, Pelvic-Ref-055, Pelvic-Ref-018, Pelvic-Ref-053, Pelvic-Ref-041, Pelvic-Ref-042