LCTSC CT and simulated CBCT, known deformation

Planning CT of the thorax (fixed, Hounsfield units) and the CBCT simulated from it (moving, normalised, no unit), deformed by a known transform the corpus adds (LCTSC through Zenodo 5035494).

24 test pairsthoraxsame patientfixed CT, moving CBCTdelivered aligned

Look at the images

Read before comparing (6)

Pair. The deposit's base pair, without its geometric augmentations: the CBCT (OS-SART reconstruction, 0.5/0.5) is simulated from its planning CT on the same cropped grid. No real cone-beam acquisition is involved.
Values. The provider scales each planning CT to 0–1; preparation restores Hounsfield units by fitting the scale to the patient's original TCIA CT: median case mean absolute residual 0.311 HU (maximum case mean 2.722 HU), at most 0.576% of voxels off by over 20 HU, chiefly near crop edges. The simulated CBCT stays on the normalised scale, which differs by patient.
Field of view. The simulated cone is empty outside its reconstructed field; preparation marks those voxels outside the evaluation mask or inside the ignore region.
Truth. The known deformation is added to the CBCT side only, and the points paired through it are exact up to interpolation. With no registration the 24 test patients score a median TRE of 1.224 mm (range 1.123–1.498 mm): the deformation is deliberately mild.
Labels. LCTSC's clinical contours as the deposit resampled them, drawn on the planning CT, not on a real CBCT. The four are scored as separate binary label sets, shown, not ranked: they overlap, and the exact reference is the points.
Overlap. These patients are LCTSC's: every test patient is in its official Test set and every development patient in its Train-side development set. A method trained on either collection must exclude this family consistently.

		Accuracy			Smoothness		Cost
#	Method	TRE mm	Change	rTRE	SDlogJ	Folds %	Time s	RAM GB	GPU GB	Images
11–3	greedyCPUapp	0.27 ±0.02	−0.97	0.0009 ±0.0001	0.023	0.00	14	1.3
21–3	deedsCPUapp	0.29 ±0.01	−0.95	0.0009 ±0.0001	0.018	0.00	21	1.1
31–3	ConvexAdam (MIND)GPUapp	0.27 ±0.02	−0.96	0.0009 ±0.0001	0.030	0.00	6	2.4	1.8
4	NiftyRegCPUapp	0.78 ±0.07	−0.46	0.0024 ±0.0002	0.050	0.01	24	0.7
	No registration	1.24 ±0.03		0.0039 ±0.0001	0.000	0.00

The order averages each method's ranks on TRE mm, SDlogJ, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 24 pairs. Two neighbours are tied when a paired Wilcoxon test on the shared pairs does not find the upper one better on TRE mm (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.

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

Measure

TRE mm	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24
greedy	0.26	0.20	0.22	0.22	0.22	0.22	0.27	0.22	0.28	0.28	0.34	0.28	0.39	0.32	0.28	0.28	0.26	0.24	0.35	0.34	0.27	0.33	0.22	0.23
deeds	0.26	0.26	0.29	0.29	0.28	0.27	0.28	0.26	0.26	0.29	0.29	0.30	0.29	0.31	0.29	0.30	0.36	0.26	0.39	0.28	0.31	0.29	0.28	0.29
ConvexAdam (MIND)	0.27	0.23	0.23	0.21	0.24	0.23	0.26	0.25	0.27	0.26	0.31	0.27	0.32	0.30	0.26	0.28	0.29	0.29	0.35	0.31	0.31	0.34	0.26	0.27
NiftyReg	0.73	0.44	0.47	0.69	0.65	0.55	0.65	0.64	0.80	0.83	1.17	0.90	0.81	0.91	1.03	0.99	0.74	0.87	0.88	0.77	0.67	0.86	0.71	0.91
No registration	1.39	1.28	1.20	1.17	1.23	1.23	1.22	1.17	1.27	1.19	1.29	1.12	1.21	1.17	1.26	1.17	1.18	1.50	1.26	1.28	1.20	1.28	1.28	1.18

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

The first test case of LCTSC CT and simulated CBCT, known deformation

The first test caseClick to look at every case.

Test pairs: All 24 patients of the provider's test archive are test cases. 10 development patients are drawn from the 20 Train patients LCTSC reserves for development, by SHA256(2026:PsCBCTThorax:<patient>) within site (4/3/3 for S1/S2/S3). No patient crosses the LCTSC/PsCBCT family partitions.
Truth: landmarks a known deformation cord by experts esophagus by experts heart by experts lungs by experts One hundred points drawn in the body inside the CBCT's field of view, paired through the inverse of the known added B-spline deformation.
Source: zenodo.org/records/5035494
doi.org/10.5281/zenodo.5035494
doi.org/10.1088/1361-6560/abe2eb
www.cancerimagingarchive.net/collection/lctsc
doi.org/10.7937/K9/TCIA.2017.3R3FVZ08
Licence: CC BY 4.0 (the Zenodo record); the planning CTs and contours it is made of are LCTSC's, CC BY 3.0

10 pairs for tuning, never tested

Train-S1-003, Train-S1-006, Train-S1-008, Train-S1-009, Train-S2-001, Train-S2-010, Train-S2-011, Train-S3-004, Train-S3-006, Train-S3-009