fidoptPelvic landmarks MR/CT (LUMC)
T2-weighted pelvic MR of a cervical cancer patient with a brachytherapy applicator in place (fixed) and her CT seven weeks earlier (moving), with landmarks placed by people on both (LUMC interobserver study).
Read before comparing (9)
- Study. One of two datasets made from the study, with PelvisLandmarksCT; each image is in one case only. Each observer saw the first MRI with its 23 reference points shown, rigidly registered to the CT, and placed the corresponding point of each on the CT in RayStation 10B.
- Truth. The fixed point is the reference placement, one person's checked by a second; the moving point is the geometric median of the observers' placements. Eight observers took part, six radiation therapy technologists and two PhD students, six or seven a patient, 5 to 7 placements a landmark after the exclusions.
- Truth error. This is the weakest of the study's settings: placements lie a median 1.5 mm from their geometric median on bone and 5.6 mm on soft tissue, some more than 50 mm away. Resampling the observers (1,000 draws) moves the geometric median a mean 3.3 mm (2.1 to 5.0 a case; 1.1 on bone, 4.9 on soft tissue).
- Exclusions. The study's own: placements of S2/S3 and S3/S4 where observers took one vertebra for the next (P06, P08, P09), and S1/S2 on P08, too close to the edge of the CT's field of view. S1/S2 and S2/S3 are not on P06's reference MRI, S1/S2 not on P08's. That leaves 21 to 23 pairs a case, 8 to 10 of them on bone.
- Alignment. The MRI and CT come 145 to 230 mm apart. The start is ANTs' rigid stage on mutual information (3.5 to 5.3 degrees), from the intensities alone, written in the moving headers; elastix placed P09 21 cm off along the body in every mode tried. The start leaves the bony landmarks 2.6 to 6.4 mm apart, where a rigid fit to them would leave 1.5 to 2.2 mm.
- Baseline. With no registration from that start: 13.2 mm (8.8 to 18.2) over all landmarks, 10.7 to 30.2 mm on soft tissue, 4.0 times the truth's own error; P08 has the least room, 9.1 mm against 2.9 mm. The TotalSegmentator labels, shown and not ranked, overlap with a Dice of 0.53 (0.44 to 0.63), HD95 22.7 mm.
- Images. The MRI is the study's first, a 1.5 T axial T2-weighted TSE with 4 mm slices over 160 mm, for brachytherapy planning; P06's is dated a week after its second. The CT, 47 to 53 days earlier, has no applicator, a mismatch the paper names among this setting's causes of spread, and is cut to the MRI's 160 mm.
- Points. Read from each scan's RTSTRUCT, they agree with the study's tables within 0.02 mm. The clinical contours in the same files, by unnamed people, are left out; the labels are TotalSegmentator's.
- Ranked apart. It misses the truth bar on its four test pairs, where a main board needs ten, and passes the two other checks: people placed the points on both images, and the start is 4.0 times the truth's own error.
| Accuracy | Smoothness | Cost | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | Dice | Change | HD95 mm | TRE mm | rTRE | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| 11–3 | deedsCPUapp | 0.681 ±0.076 | +0.152 | 18.07 ±10.88 | 8.25 ±7.21 | 0.0228 ±0.0199 | 0.081 | 0.00 | 24 | 1.4 | ||
| 21–3 | ConvexAdam (MIND)GPUapp | 0.604 ±0.111 | +0.074 | 21.98 ±12.14 | 13.60 ±5.69 | 0.0375 ±0.0157 | 0.090 | 0.20 | 11 | 2.4 | 2.2 | |
| 31–3 | NiftyRegCPUapp | 0.597 ±0.121 | +0.067 | 20.70 ±12.48 | 13.74 ±8.04 | 0.0379 ±0.0222 | 0.103 | 0.57 | 29 | 1.3 | ||
| No registration | 0.530 ±0.126 | 22.71 ±12.25 | 13.22 ±7.84 | 0.0365 ±0.0216 | 0.000 | 0.00 | ||||||
The order averages each method's ranks on Dice, HD95 mm, TRE mm, SDlogJ, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 4 pairs. Two neighbours are tied when a paired Wilcoxon test on the shared pairs does not find the upper one better on Dice (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.
Dice: how much the organ contours overlap, 1 is perfect. HD95: distance between the contours' surfaces. 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
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
- The four MRI-CT pairs of P06 to P09, all test pairs, one a patient; none is left for development. P05's pair is left out: its CT is PelvisLandmarksCT/P05's full-bladder CT. The MRI-MRI pairs are not used: they share the fixed image and start only 2.4 to 2.7 times the truth's own error apart.
- Truth
- landmarks by experts contours by a program TotalSegmentator run by us on each image (datasets/segment.py total_mr,total), task total_mr on the MRI and task total on the CT, reduced to the structures both tasks name, in total_mr's ids: the CT's lung lobes make the two lungs and its S1 joins its sacrum, the vertebrae and discs are left out; its prostate is taken out (prepare.py --labels), these women having none; a secondary truth, graded automatic: shown, not ranked, the landmarks rank 23 anatomical landmarks defined for pelvic MRI, 10 of them on bone: on the MRI the reference point, placed by an experienced radiation therapy technologist and verified by a gynaecological radiation oncologist; on the CT the geometric median of the 5 to 7 observers' placements, the study's own consensus. 21 to 23 a pair
- Source
- doi.org/10.4121/b49aadc1-e435-4d67-a7d2-460d9ed02558.v1
data.4tu.nl/datasets/b49aadc1-e435-4d67-a7d2-460d9ed02558/1
doi.org/10.1016/j.radonc.2026.111430
ir.cwi.nl/pub/36285 - Licence
- CC BY-NC-ND 4.0