fidoptSynthRAD2023, CBCT to CT, pelvis
Cone-beam CT of one patient's pelvis and the CT to synthesise from it, the CT brought onto the CBCT's grid by the challenge (SynthRAD2023, Task 2).
Read before comparing (5)
- Cases. The challenge's pair as delivered, copied byte for byte: the CBCT, the CT the organisers registered onto its grid, and their body mask. The synthetic image is scored inside that mask, as the challenge scores it.
- Alignment. How the organisers registered the CT to the CBCT decides how far a perfect synthesis can match it: what the registration left misaligned counts as error for every method.
- Baseline. Inside the mask on the 14 test cases, with KonfAI's criteria at CT's range of 4095, a constant at the target's median scores an MAE of 237 HU (180 to 298), a PSNR of 20.0 dB (17.8 to 22.0) and an SSIM of 0.30 (0.15 to 0.39).
- CBCT values. The CBCT taken as the CT is no floor: its values are not Hounsfield units at every centre. Its MAE on the test cases is 616 to 758 HU at 2PA, 259 to 299 at 2PB and 169 to 210 at 2PC.
- Overlap. No dataset outside the eight SynthRAD2023 sets draws on its record. The provider says its two tasks 'do not necessarily contain the same patients', so every drawn case's CT was registered rigidly to every CT of SynthRAD_MRCT's 34 and SynthRAD_MRCT_Syn's cases: they correlate at most 0.88 inside the bodies, where one CT on two grids reads 0.98 or more.
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 | Anatomy Dice | Anatomy HD95 mm | Time s | RAM GB | GPU GB | Machine | Images |
| nnsyn · CBCT to CT, abdomenGPUapptrained on CC-BY-NC-4.0 dataolder run | 100.5 ±16.5 | 25.80 ±1.16 | 0.719 ±0.030 | 0.766 ±0.026 | 7.96 ±1.19 | 4.0 | 2.6 | 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. Anatomy Dice and HD95: TotalSegmentator run on the synthetic and on the target image, the structures compared; shown, not ranked.
Case by case
| MAE | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 |
|---|
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
- Test cases
- 14 test patients drawn with seed 2026 by centre (2PA, 2PB, 2PC) in proportion, then 13 development patients, the last of the same seeded order from the centre with the most left. The pool is the 146 of SynthRAD2023 Task 2's 180 pelvis training patients outside SynthRAD_CBCTCT's 18 test and 16 development patients, whose CT, the target here, is that registration set's public fixed image.
- Truth
- target image as released the patient's CT as the challenge delivers it, registered to the CBCT by the organisers, in Hounsfield units
- Licence
- CC BY-NC 4.0
13 cases for tuning, never tested
pelvis_2PA041, pelvis_2PB008, pelvis_2PC016, pelvis_2PA035, pelvis_2PB104, pelvis_2PC029, pelvis_2PA064, pelvis_2PB049, pelvis_2PC012, pelvis_2PA018, pelvis_2PB111, pelvis_2PC020, pelvis_2PA043