fidopt Interactive view

QUADRA_HC, whole-body CT test-retest

Low-dose whole-body CT, vertex to thighs, of one healthy adult scanned twice on one PET/CT scanner weeks apart: first session fixed, second moving (QUADRA_HC, Medical University of Vienna).

10 test pairswhole bodysame patientfixed CT, moving CTaligned once by the benchmark
Read before comparing (7)
  • Cohort. 48 healthy adults, 25 women and 23 men aged 19 to 59, each with two FDG PET/CT sessions on a Siemens Biograph Vision Quadra, 38 days apart on average (22 to 98). The CTs are in Hounsfield units, cut to the box of the body and the table; the PET is not used.
  • Truth. Ranked apart: MOOSE 3.0.13 labelled each CT alone and nobody corrected the maps. The Zenodo record mentions only 'basic sanity checks (laterality, presence of volumes, gross structure)', and the data descriptor says the maps were not verified one by one. A method built on MOOSE or TotalSegmentator labels is checked against a close relative's view of the anatomy.
  • Scoring. A structure a map lacks on one image is not scored on that pair: an image holds 116 to 120 of the 132 structures on the 20 cases built.
  • Alignment. Both sessions keep their voxels and grids. The moving image's and labels' headers carry one elastix translation from mutual information inside each body, no annotation used.
  • Baseline. With no registration the ten test pairs start at a mean Dice of 0.55 (0.33 to 0.70) over the structures both maps hold, and an HD95 of 12.6 mm (6.5 to 20.1). Small and mobile structures score lowest (adrenal glands, gallbladder, bladder), as do the arms, which lie otherwise at the second session.
  • Overlap. MultimodalHC_PETCT_Syn and MultimodalHC_MRCT share the scanner model and CT grid but not the cohort: 100 Danish blood donors scanned at Rigshospitalet, Copenhagen. Neither provider names the other's cohort, and no subject list links them.
  • Outside numbers. None: no registration result on these pairs is published. Every subject is public with both CTs and their labels.
AccuracySmoothnessCost
#MethodDiceChangeHD95 mmSDlogJFolds %Time sRAM GBGPU GBImages
11–3NiftyRegCPUapp0.840 ±0.015+0.2887.22 ±2.710.0300.013165.2
21–3deedsCPUapp0.843 ±0.016+0.2917.07 ±2.960.0850.0319510.8
31–3ConvexAdam (MIND)GPUapp0.838 ±0.019+0.2877.73 ±2.940.0900.18265.414.8
33–5greedyCPUapp0.813 ±0.021+0.2618.40 ±2.700.0300.0019410.8
53–5DemonsCPUapp0.811 ±0.023+0.2597.96 ±2.650.0520.00457.7
No registration0.552 ±0.07912.60 ±2.910.0000.00

The order averages each method's ranks on Dice, HD95 mm, SDlogJ, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 10 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. 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

The first test case of QUADRA_HC, whole-body CT test-retest
The first test caseClick to look at every case.
Test pairs
10 test subjects drawn with seed 2026 by sex among the dataset's 48 (5 women, 5 men), then 10 development subjects (6 and 4). One pair a subject, so no subject is in both sets.
Truth
contours by a program the provider's MOOSE 3.0.13 maps of each CT, made by a network on each image alone: seven of its eight files put into one map, 132 structures (19 organs, 13 of the heart and great vessels, 4 of the digestive tract, 10 muscles, 31 bones of the limbs and skull, 26 ribs and the sternum, 25 vertebrae with the hips and the sacrum)
Licence
CC BY 4.0
10 pairs for tuning, never tested

QUADRA_HC_024, QUADRA_HC_019, QUADRA_HC_021, QUADRA_HC_036, QUADRA_HC_043, QUADRA_HC_020, QUADRA_HC_030, QUADRA_HC_011, QUADRA_HC_045, QUADRA_HC_015