fidoptRHUH-GBM, glioblastoma before surgery
Pre-operative brain MR of a patient with a glioblastoma later totally or near totally resected at Río Hortega University Hospital, Valladolid: T1, post-contrast T1, T2 and FLAIR as one four-channel image on SRI24, skull-stripped, with necrosis, oedema and enhancing tumour labelled (RHUH-GBM, TCIA).
10 test casesbrainimage MR
Read before comparing (7)
- Pool. 40 consecutive adults operated on from 2018 to 2022 for a WHO grade 4 astrocytoma with a total or near-total resection, each scanned at 1.5 T before surgery, early after it and at recurrence. Only the scan before surgery is used, one a patient. RHUH-0016 (not drawn) and RHUH-0024 (development) had surgery and chemoradiation before, so their scan before surgery shows a recurrence.
- Images. The provider registered each series to the post-contrast T1 on SRI24 with FLIRT, resampled it to 1 mm, skull-stripped it with SynthStrip and normalised its intensities: values without unit, about -10 to 12, the background a negative constant. RHUH-0028 (development) lies on a 230 x 230 x 138 grid of its own. On some cases the brain sits turned on the SRI24 grid, as test case RHUH-0001's by about 25 degrees in the axial plane. The ADC map is not given.
- Truth. DeepMedic's labels, corrected by hand by two neurosurgeons specialised in neuroimaging; no agreement between readers is published and how much of DeepMedic's output remains is not said. Graded semi-automatic. On the cases looked at, 3 reads brightest on the post-contrast T1, 1 on the T2 and 2 on the FLAIR.
- Labels. 1 the necrosis, 2 the peritumoral signal change with any non-enhancing tumour, 3 the enhancing tumour, as BraTS 2021 numbers them with 4 for 3; named as in BraTSAfrica. Over the test cases the necrosis holds 0.16 to 73 ml (median 4.2), the oedema 17 to 93 ml, the enhancing tumour 3.5 to 42 ml.
- Start. Over the 10 test cases a whole-image mask scores a mean Dice over the three labels of 0.006 at the median, an Otsu mask of the T1 0.035 (0.010 to 0.059).
- Overlap. The patients are Río Hortega's, operated on from 2018; no other descriptor names the hospital, Valladolid or the collection. BraTS 2021's contributors include no Spanish centre, and the corpus's other glioma sets come from other named institutions (UT Southwestern, Missouri, UC San Diego, Nigeria, Brigham, the paediatric consortia). A check by images did not count: on SRI24 the same patient's scans at two timepoints correlate voxel by voxel at a median of 0.20, no better than two patients (up to 0.50), so the highest correlation with the 266 built cases of BraTS2021, BraTSAfrica, BraTSPEDs, BraTS_MEN, GliomaUTSWMR and MUGliomaPostMR, 0.47, says nothing.
- Outside numbers. None: the collection 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 patients drawn with seed 2026 among the 40, then 10 development patients, the last of the same seeded order. A case is a patient's scan before surgery; two centres, not named a patient, so no strata.
- Truth
- contours by a program, verified by people the dataset's three labels, 1 the necrosis, 2 the peritumoral signal change (oedema and non-enhancing tumour), 3 the enhancing tumour: DeepMedic's segmentation of the preprocessed images, reviewed and corrected by hand by two neurosurgeons specialised in neuroimaging (Cepeda et al. 2023)
- Source
- www.cancerimagingarchive.net/collection/rhuh-gbm
doi.org/10.7937/4545-c905
doi.org/10.1016/j.dib.2023.109617
www.cancerimagingarchive.net/wp-content/uploads/clinical_data_TCIA_RHUH-GBM.csv
github.com/smcch/RHUH-GBM-dataset-MRI-preprocessing - Licence
- CC BY 4.0
10 cases for tuning, never tested
RHUH-0009, RHUH-0024, RHUH-0019, RHUH-0029, RHUH-0023, RHUH-0033, RHUH-0018, RHUH-0012, RHUH-0026, RHUH-0028