fidoptSynthStrip
One whole-head MR of a subject, adult, glioblastoma patient or infant, in one of nine contrasts and resolutions, with its brain mask (SynthStrip dataset, v1.5).
15 test casesbrainimage MRresearch only
Read before comparing (7)
- Pool. The deposit's 622 scans of 221 subjects, less IXI's 50 subjects (their images are IXI_TransMorph's and IXI_T2PD's) and CERMEP-IDB-MRXFDG's 20 (their PET and CT need a request form): FSM's 38 in-house subjects (T1 MPRAGE, T2 SPACE, PD FLASH, quantitative T1 MP2RAGE), ASL's 43 HCP-Aging subjects (T1 MPRAGE, PCASL EPI at 3.4 by 3.4 by 5 mm), QIN GBM Treatment Response's 54 patients (thick-slice 2D T1, T2, FLAIR) and 16 infants from Boston Children's Hospital (T1 MPRAGE).
- Cases. The 15 test images: 2 PCASL EPI and 2 T1 (ASL), 3 T2 and 1 PD (FSM), 2 infant T1, 2 FLAIR, 2 T1 and 1 T2 (QIN). The images are the source datasets' files unmodified, in their own units and orientations; the QIN slices are 5 or 6 mm thick.
- Truth. Graded semi-automatic: the masks start from tools' votes, then people refined them on every subject's T1 and fixed the propagations that failed; a mask on another contrast keeps the T1's outline where the registration was good. The labels.nii.gz label maps of the adult MPRAGE scans are a model's fit and are not used.
- Start. A mask of the whole image scores a Dice of 0.18 at the median of the test images, an Otsu mask 0.66 (0.94 on the EPI asl_epi_132, whose field holds little but brain; 0 on qin_t2_38). TotalSegmentator 2.18.0's total_mr brain scores 0.979 on fsm_t2_07tz, 0.965 on qin_t1_12 and 0.962 on asl_epi_132, but 0.790 on infant_t1_19: the room lies in the infants, the thick slices and the EPI.
- Overlap. IXI's subjects are left out; no other dataset of the corpus names QIN GBM Treatment Response, HCP-Aging or the Boston infant cohort. SynthStrip itself was trained on images synthesised from the label maps of 80 other subjects (Buckner40, 30 HCP-Aging adults, 10 Boston infants), held out of this set.
- Licence. The deposit's CC BY 4.0 does not settle two sources. TCIA served QIN GBM Treatment Response as limited access in 2016, under CC BY 3.0 in 2021, and now under the NIH Controlled Data Access Policy, faces being rebuildable from these head scans; HCP-Aging shares its imaging only through the NIMH Data Archive under a data use certification. Hence ask: the FSM and Infant subjects alone would carry only the deposit's CC BY 4.0.
- Outside numbers. None on these cases: the paper reports SynthStrip's Dice per subset over its own test scans, 95.2 on the PCASL EPI, 96.1 on the infants, up to 97.7.
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
- 15 test images drawn with seed 2026 among the 137 subjects outside development, the four cohorts in proportion, and 14 development images. A subject is one case: one of its scans, picked by a seeded hash so the contrasts mix. The paper's validation and test lists split scans, not subjects, and are not used.
- Truth
- contours by a program, verified by people the deposit's binary brain mask on the image's grid: the majority vote of five classical skull-stripping tools on the subject's T1-weighted scan, refined by hand, carried to the subject's other scans by a robust rigid registration, poor alignments refined by hand; in v1.5 the largest connected part with its holes filled
- Source
- w3id.org/synthstrip
surfer.nmr.mgh.harvard.edu/docs/synthstrip
surfer.nmr.mgh.harvard.edu/docs/synthstrip/data/README
doi.org/10.1016/j.neuroimage.2022.119474
arxiv.org/abs/2203.09974 - Licence
- CC BY 4.0 by the deposit's README, but its QIN images come from a TCIA collection, CC BY 3.0 when taken, that TCIA now keeps under NIH controlled access, and its ASL images are HCP-Aging scans, whose imaging the project shares only through the NIMH Data Archive
14 cases for tuning, never tested
qin_flair_43, asl_t1_118, fsm_qt1_21em, qin_t1_47, infant_t1_23, asl_t1_127, fsm_pd_50ww, qin_t1_57, asl_epi_119, qin_t2_46, fsm_pd_10cv, asl_t1_124, qin_t1_31, fsm_pd_76cw