fidopt Interactive view
Chest CT, segmentation

AeroPath

A contrast-enhanced chest CT angiography of a patient examined for lung cancer, most with disease that deforms the airways, with the airway tree labelled (AeroPath).

10 test caseslungimage CT
Look at the images
Read before comparing (8)
  • Pool. 27 CT angiography scans of 27 patients undergoing tests for lung cancer at St. Olavs hospital, Trondheim (Støverud et al., PLOS ONE 2024). Most have large tumours or collapsed segments partly blocking the airways of at least one lobe; others a dilated trachea or oesophagus, emphysema or sarcoidosis.
  • Images. The CT, in Hounsfield units, and its airway mask are copied byte for byte. They have 0.56 to 0.82 mm in plane and slices 0.5 to 1.25 mm apart; the paper gives 0.68 to 0.76 mm in plane, but the files hold the wider range.
  • Truth. Region growing in FAST, or grow-cut in 3D Slicer where it leaked, drew the trachea and first branches, and BronchiNet, a network, the peripheral airways. A trained engineer supervised by a pulmonologist corrected the tree by hand, and a pulmonologist checked every case. Graded semi-automatic: as the paper says of its own results, a method close to FAST or BronchiNet may be favoured.
  • Coverage. The tree holds 128 ± 56 branches a scan, against 208 ± 74 in ATM'22's non-contrast training scans: disease hides branches, and the labels were made for planning bronchoscopy, not to reach the smallest airways. On a few scans the paper's network found branches the labels lack; such branches count against a method.
  • Lungs. The archive's lung masks stay out of the truth: neither the paper nor the README says how they were made or whether anybody checked them. On test case 3 they match TotalSegmentator's five lobes with a Dice of 0.97.
  • TotalSegmentator. Version 2.18's lung_airways class reaches a Dice of 0.81 on test case 3's airway tree; the paper reports 0.82 ± 0.03 for an earlier version over the 27 scans, the trachea clipped.
  • Outside numbers. None compare. The paper clips the trachea at the top of the lungs and uses ATM'22's measures; here the whole tree is scored by Dice, HD95 and surface Dice, which see a missed peripheral branch only by its few voxels.
  • Overlap. The scans are this dataset's own: no other dataset of the corpus holds them.
AccuracyCost
#MethodDiceSurface DiceHD95 mmTime sRAM GBGPU GBImages
1KonfAI segmentation app · TotalSegmentator (lung_vessels), the authors' packageGPUapp0.825 ±0.0230.837 ±0.03313.49 ±3.6655.64.9

The order averages each method's ranks on Dice, Surface Dice, HD95 mm, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 10 cases. Two neighbours are tied when a paired Wilcoxon test on the shared cases 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. Surface Dice: the share of the contours' surfaces within a tolerance of each other, 1 is perfect. HD95: distance between the contours' surfaces.

Case by case

Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.

About the data

The first test case of AeroPath
The first test caseClick to look at every case.
Test cases
10 test CTs drawn with seed 2026 among the 17 outside development, and 10 development CTs, out of the dataset's 27. One CT a patient, so a patient is a test or a development case, never both.
Truth
contours by a program, verified by people the airway tree, trachea and bronchi: the dataset's airway masks, grown by programs, corrected by hand by a trained engineer supervised by a pulmonologist, and checked case by case by a pulmonologist
Licence
CC BY 4.0
10 cases for tuning, never tested

20, 1, 23, 5, 12, 2, 26, 6, 8, 13