fidoptDigital Hand Atlas radiographs
Left-hand and wrist radiographs of two children, 2D films digitised by the USC Image Processing and Informatics Lab, with 37 landmarks on the fingertips, finger joints and wrist placed by Payer et al. on each (Digital Hand Atlas).
Read before comparing (9)
- Pool. The atlas holds 1,390 left-hand radiographs of healthy children of four ethnic groups, both sexes, newborn to 18, taken at Children's Hospital Los Angeles from 1993 to 2006. Payer et al. placed 37 landmarks on 895 of them, all aged 9 to 19, and the pairs come from those.
- Children. The atlas followed some children for years: 171 of the 895 images share a birth date, sex and ethnic group with an earlier one, their heights growing, and two pairs (4195 and 6046, 5400 and 6050) are one examination digitised twice. Each child keeps its first image in the draw's order; twins of one birth date count as one child. Two more films are in the pool twice under headers that disagree (4724 and 6021, 4520 and 4521: one hand on the images); no drawn pair holds them.
- Ages. The pairs are drawn across ages, as the other radiograph sets: the two children of a test pair differ by 2.8 years in the median, at most 8.1, so size, epiphyses and carpal ossification differ between the hands.
- Images. Films digitised by the lab, 16-bit (12-bit on 14 test images), 1056 to 2081 pixels wide, as delivered with their side markers and tags. 166 of the 895 headers give 0.1 mm a pixel, the others nothing, so the unit is the pixel: at about 0.1 mm, 10 pixels are a millimetre. The DICOM laterality says R on images of the left hand. No mask.
- Truth. Payer et al. placed the 37 points by hand: the fingertips, the joints of the five fingers and points of the carpus and wrist. They report no agreement between readers; their own evaluation scales distances by taking two wrist points as 50 mm apart.
- Alignment. Each image keeps its own grid at the origin. The moving image's header carries one translation from elastix mutual information, the best of several starts, no annotation used. A rigid fit of the points turns the hands by 2.2 degrees in the median, at most 12.7, so only the translation is kept.
- Baseline. With no registration the 30 test pairs score a TRE of 130 pixels (49 to 263), rTRE 0.048, and a 95th percentile of 263 pixels on average, against 40 pixels after an affine fit on the points.
- Coverage. The points lie on the bones of the hand and wrist: nothing scores the soft tissue or the forearm.
- Outside numbers. None: the atlas's papers and Payer et al. rank bone-age estimation and landmark detection, and the pairing is ours.
No method has a result on every pair of this dataset yet.
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
- 30 test pairs drawn with seed 2026 among the 362 pairs, and 30 development pairs, the last of the same seeded order. The 724 children of Payer's annotated images, one image each, are put in the order of sha256("2026:DigitalHandAtlas:<image>") and paired two by two, the first fixed. One hospital and one digitiser: the draw is not stratified.
- Truth
- landmarks by experts the 37 landmarks Payer et al. placed by hand on each image: the fingertips, the joints of the five fingers and points of the carpal bones and the wrist (hand_xray/hand_xray_dataset/setup/all.csv)
- Source
- ipilab.usc.edu/computer-aided-bone-age-assessment-of-children-using-a-digital-hand-atlas-2
github.com/christianpayer/MedicalDataAugmentationTool-HeatmapRegression
doi.org/10.1007/978-3-319-46723-8_27 - Licence
- images: USC Image Processing and Informatics Lab, 'for research and education purpose only', no licence; landmarks: GPL-3.0
30 pairs for tuning, never tested
5094_5910, 5099_5359, 5159_4721, 3280_7101, 7021_4121, 4479_7155, 7273_5725, 6154_5267, 5361_5084, 5204_5123, 5167_6132, 3157_7118, 4383_3188, 5383_3161, 5389_5706, 3769_3236, 4488_4360, 3148_7251, 5284_5354, 3240_5533, 5003_6008, 5161_5526, 4503_7009, 5277_7018, 7022_4206, 7116_6108, 5174_6115, 7266_5058, 3281_3196, 5218_5338