fidopt Interactive view
Lesions and tumours, segmentation

BHSD brain haemorrhage CT

Non-contrast head CT of one patient with intracranial haemorrhage from the RSNA Intracranial Hemorrhage Detection data, with five kinds of bleed drawn by medical imaging experts and reviewed by a radiologist (BHSD, test split).

95 test casesbrainimage CTresearch only
Look at the images
Read before comparing (7)
  • Cases. Non-contrast head CTs of the RSNA Intracranial Hemorrhage Detection data that the authors rebuilt into volumes, every one with a bleed: 0.39 to 0.63 mm pixels, slices 2.5 to 6.4 mm apart, 24 to 60 slices; most volumes are tilted with the gantry, as their direction matrix says. Hounsfield units as delivered, the padding around the reconstructed circle at -2048 or -3024 kept. No mask.
  • Labels. Five kinds of bleed, each present on a part of the test cases: intraparenchymal 62, subarachnoid 50, intraventricular 54, subdural 34, epidural 13. A case may hold several; a structure absent from a case is not scored there.
  • Truth. Two medical imaging experts and radiology residents with over a year of reading head CT drew the bleeds in ITK-SNAP, and a board-certified radiologist reviewed them (Wu et al. 2023). One delineation a case; no agreement between readers is published.
  • Held out. ID_56faaf6f_ID_dd8f688dff, a test case, is held out: its file gives 0.5 mm between its 48 slices, so the head is 24 mm tall; the true spacing is not given.
  • Start. A mask of the whole image scores a Dice of 0.001 at the median over the test cases and bleeds, an Otsu mask 0.002 (0 to 0.078). TotalSegmentator 2.18.0's cerebral_bleed, one class, overlaps the union of the bleeds at 0.881 on ID_0237f3c9_ID_40015688b9 and 0.851 on ID_02b882cc_ID_a4892e60ae, with no kinds.
  • No derivatives. CC BY-NC-ND 4.0: the cases are built and scored on the reference machine only and never uploaded; nothing computed from them is shared.
  • Overlap. No other dataset of the corpus holds RSNA Intracranial Hemorrhage Detection images; the provider's two splits share no patient (its paper, checked on the patient identifiers in the file names).

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
The provider's test split (dataset.json of its repository), 96 head CTs, less one: 95 test cases. 10 development cases drawn with seed 2026 from its 96 training cases, one study a patient: patient ID_2db7ee14 has two training studies, ID_2db7ee14_ID_f27e38fdfd held out.
Truth
contours by experts the deposit's label map on the CT: drawn in ITK-SNAP by two medical imaging experts and radiology residents with over a year of reading head CT, each case reviewed by a board-certified radiologist (Wu et al. 2023)
Licence
CC BY-NC-ND 4.0
10 cases for tuning, never tested

ID_ba3a77dd_ID_70715be365, ID_96967deb_ID_5d782e9379, ID_b79a2d4a_ID_710cdb6530, ID_8d326f80_ID_87b6843dde, ID_73cae446_ID_852cb06998, ID_089596a9_ID_ea71b57820, ID_05331dc6_ID_47213b7e2e, ID_4ec93c83_ID_b76a4106f8, ID_346de481_ID_2e8fea0d9a, ID_50b54af0_ID_fbeeed4370