fidoptNFBS brain masks
Defaced T1-weighted MR of one adult of the Enhanced Nathan Kline Institute Rockland Sample's neurofeedback study, with its brain mask from BEaST checked by eye and edited by hand where needed (Neurofeedback Skull-stripped repository).
12 test casesbrainimage MRresearch only
Read before comparing (6)
- Cases. One defaced T1-weighted MPRAGE a participant, 1 mm isotropic, 256 by 256 by 192, as delivered, of 125 adults aged 21 to 45 with a range of clinical and subclinical psychiatric symptoms, scanned for the Enhanced NKI Rockland Sample's neurofeedback study. No mask.
- Truth. BEaST stripped every scan; the authors checked each result by eye and edited it by hand in Freeview where needed, 1 to 8 hours a scan, then added the edited masks to BEaST's library and stripped the rest again, about 85 of the 125 edited in all (Puccio et al. 2016). Which masks were edited is not said: graded semi-automatic. The brain is everything inside the dura: skull, scalp, fat, muscle, dura and external vessels and nerves are out.
- Masks. The repository stores the masks as floats; on each, 8 to 171 voxels hold a fraction, a resampling's trace, and those of 0.5 or more count as brain.
- Start. Labelling the whole image as brain scores a Dice of 0.190 at the median of the test cases (0.161 to 0.214), an Otsu mask 0.539 (0.477 to 0.638). TotalSegmentator 2.18.0's total_mr brain scores 0.917 on A00043998, 0.911 on A00051676 and 0.907 on A00052340.
- Overlap. Mindboggle101 holds NKI-RS-22 and NKI-TRT-20 brains of the earlier NKI Rockland release, with no identifier to link. Each brain was put in MNI152 space by an affine fit and correlated with Mindboggle-101's 42 NKI brains: A00028185 reaches 0.815 with NKI-RS-22-14 and NKI-TRT-20-14 (one person), where any other pair stays at 0.623 or less and two NFBS participants at 0.728 or less; the two people Mindboggle-101 holds twice reach 0.987 and 0.989. A00028185 is held out. LUMIR's OpenBHB brains, which include CoRR's NKI scans, were matched the same way: none of its 120 built brains reaches more than 0.616 with the 23 built NFBS brains, below two other LUMIR brains (0.667) or two NFBS brains (0.716).
- Room. Brain extraction on clean 1 mm T1 scans is close to solved: the board separates methods by their edges more than by their overlap. SynthStripMR covers other contrasts and resolutions.
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
- 12 test participants drawn with seed 2026 and 11 development participants, one scan each, from the repository's 125 less A00028185, who appears to be in Mindboggle-101 (Overlap). One scanner and protocol: the draw is not stratified.
- Truth
- contours by a program, verified by people the repository's brain mask on the T1: BEaST's output checked by eye, edited by hand in Freeview where needed, about 85 of the 125 masks, the edited masks added to BEaST's library to strip the rest; skull, scalp, fat, muscle, dura and external vessels and nerves removed (Puccio et al. 2016)
- Source
- preprocessed-connectomes-project.org/NFB_skullstripped
doi.org/10.1186/s13742-016-0150-5
fcon_1000.projects.nitrc.org/indi/enhanced - Licence
- none stated on the repository's page or in the archive; the Enhanced NKI Rockland Sample images are shared by INDI
11 cases for tuning, never tested
A00060169, A00061711, A00062288, A00061709, A00035840, A00038642, A00045590, A00059845, A00062210, A00037511, A00056452