fidoptIXI T2 to PD
T2-weighted (source) and proton-density-weighted (target) MR of the head of one healthy subject, the two echoes of one acquisition (IXI).
Read before comparing (8)
- Alignment. The T2 and the PD are two echoes of one acquisition, on one grid as delivered: nothing is registered or resampled. The scanner pages give both one repetition time and echo train at Guy's and Hammersmith, the echo times 100 and 8 ms, and the Institute of Psychiatry's headers one repetition time. Size, spacing, origin and direction match on 577 of the 578 subjects.
- Scanners. Healthy subjects scanned at Guy's (Philips 1.5 T), Hammersmith (Philips 3 T) and the Institute of Psychiatry (GE).
- Values. The scanners' own 16-bit units, times the header's scale factor where there is one. 102 subjects carry one, among them three test subjects (IXI021-Guys-0703, IXI110-Guys-0733, IXI138-Guys-0746) and 7 development subjects, which read about three times brighter. The scale varies from scan to scan anyway, shared by a scan's two echoes. The MAE is in each scan's units, so its mean weighs the brighter scans most.
- Range. The PSNR and SSIM use one range for the dataset, 0 to 2789: the 99.9th percentile of the PD inside the body, pooled over the ten test subjects drawn first. The subjects' own run from 652 to 3722.
- Mask. None: the measures are taken on the PD's Otsu foreground, the head without what is dark in the PD (bone, air, some vessels), 23% to 37% of the image.
- Target. A real acquisition with its noise and artefacts, such as stripes where a subject moved, the same in the T2: no method reaches an MAE of 0.
- Overlap. IXI_TransMorph renames its subjects; the release's demographic table, matched to IXI.xls, names them, and IXI_TransMorph leaves out every subject that is one of these 60 or that the table does not name.
- Outside numbers. Papers on IXI T2-to-PD synthesis choose their own subjects and slices and normalise the intensities: no published number is scored as here. IXI is public and widely used to train synthesis networks, so a network may have seen a test subject and its PD.
No method has a result on every case of this dataset yet.
| Accuracy | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| # | Method | MAE | Change | PSNR dB | SSIM | Time s | RAM GB | GPU GB | Images |
| Source as given | 370.2 ±76.0 | 18.13 ±1.57 | 0.616 ±0.010 | ||||||
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
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
- 30 test subjects drawn with seed 2026 by hospital in proportion among the 577 with a T2 and a PD on one grid: 16 Guy's, 10 Hammersmith, 4 Institute of Psychiatry. 30 development subjects, the last of the same seeded order, those the test draw rests on kept, each hospital in proportion: 17 Guy's, 9 Hammersmith, 4 Institute of Psychiatry, none an IXI_TransMorph subject. IXI014-HH-1236 is left out: its T2 and PD are two acquisitions on two grids.
- Truth
- target image as released the subject's PD-weighted image, the other echo of the T2's acquisition, as delivered
- Source
- brain-development.org/ixi-dataset
brain-development.org/scanner-philips-medical-systems-gyroscan-intera-1-5t
brain-development.org/scanner-philips-medical-systems-intera-3t - Licence
- CC BY-SA 3.0
30 cases for tuning, never tested
IXI121-Guys-0772, IXI641-Guys-1105, IXI494-Guys-1008, IXI189-Guys-0799, IXI582-Guys-1127, IXI308-Guys-0884, IXI179-Guys-0777, IXI229-Guys-0980, IXI088-Guys-0758, IXI069-Guys-0769, IXI364-Guys-0920, IXI170-Guys-0843, IXI025-Guys-0852, IXI420-Guys-1028, IXI523-Guys-1056, IXI122-Guys-0773, IXI134-Guys-0780, IXI253-HH-1694, IXI427-IOP-1012, IXI049-HH-1358, IXI631-HH-2651, IXI464-IOP-1029, IXI239-HH-2296, IXI354-HH-2024, IXI357-HH-2076, IXI395-IOP-0969, IXI438-HH-2155, IXI159-HH-1549, IXI423-IOP-0974, IXI567-HH-2536