fidopt Interactive view
Chest CT, segmentation

MIDRC-RICORD-1A

One chest CT of an adult with COVID-19, from four sites in four countries, with the infectious opacities outlined slice by slice by a thoracic radiologist (MIDRC-RICORD-1A, RSNA and the Society of Thoracic Radiology).

10 test caseslungimage CTresearch only
Look at the images
Read before comparing (8)
  • Pool. 120 chest CT studies of 110 COVID-19 positive adults from four sites, named by pseudonymous site ids (419639, 440808, 660042, SITE2); 118 studies are outlined. A patient is one case: the earliest outlined study of the eight patients with several. Left out: two patients without outlines and four whose outlines hold only non-infectious findings.
  • Truth. Six subspecialist thoracic radiologists outlined the findings slice by slice under a schema agreed with the STR and EuSoMII (Tsai et al. 2021), only opacities and nodules over 1 cm, so a smaller opacity is background; 86 of the pool's 104 series hold one reader's polygons, 18 two readers'. No agreement between readers is published. Graded expert. Some outlines skip a slice now and then: on 419639-001155 (slices 1.25 mm apart) the coronal view shows upper-lobe opacities missing on single slices.
  • Labels. lung_covid_infiltrate merges the three infectious findings, as COVIDLungCT's infection. The non-infectious findings (atelectasis, nodule or mass, other opacity) are taken out of the truth and of the prediction (ignore.nii.gz), where no infectious finding lies: present on 3 test CTs, 1.9 mL at most. The surface Dice uses 3 mm, as COVIDLungCT.
  • Cases. The test infiltrates hold 0.9 to 2312 mL (median 451 mL); 660042-000100 (0.9 mL in 5 polygons) and SITE2-000198 (1.7 mL of tree-in-bud and a cavity) are small, where a single missed or extra blob moves the Dice from 0 to 1.
  • Images. The outlined series as delivered, in Hounsfield units, slices 1.25 to 3 mm apart, pixels 0.61 to 0.88 mm on the test CTs; three read -3024 outside the reconstructed circle. prepare.py stops on a series whose slices are not evenly spaced.
  • Start. A mask of the whole image scores a Dice of 0.022 at the median of the test CTs (0 to 0.103), an Otsu mask 0.032. TotalSegmentator 2.18.0's five lung lobes as a mask score 0.705 on 440808-000008, whose opacities fill 66% of the lungs, and 0.359 on SITE2-000001 (25%).
  • Overlap. No other dataset of the corpus names MIDRC or RICORD; COVIDLungCT's CTs come from Coronacases and Radiopaedia. RICORD 1b (COVID-negative CTs) and 1c (radiographs) are not used.
  • Outside numbers. None: RICORD has no challenge and no official split.

No method has a result on every case of this dataset yet.

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

Test cases
10 test CTs drawn with seed 2026 among the 93 patients outside development, the four sites in proportion, and 10 development CTs. A case is a patient's earliest outlined study; the 104 patients whose outlines hold an infectious finding make the pool, less SITE2-000051, whose series holds two different images at ten slice positions. RICORD publishes no test set.
Truth
contours by experts the polygons of the findings 'Infectious opacity', 'Infectious TIB/micronodules' and 'Infectious cavity' on the outlined axial series, filled with their edge: drawn by hand on each slice in MD.ai by one of six thoracic radiologists (6 years of experience on average), one reader a series for most; the non-infectious findings outlined beside them (nodule or mass, atelectasis, other opacity) are the ignore region
Licence
CC BY-NC 4.0
10 cases for tuning, never tested

SITE2-000024, 440808-000019, 660042-000063, SITE2-000016, 419639-001336, SITE2-000007, SITE2-000038, 440808-000016, 660042-000107, SITE2-000174