Full details are now published. Please cite this work when using the data.
Phillip G.D. Ward, Nicholas J. Ferris, Parnesh Raniga, David L. Dowe, Amanda C.L. Ng, David G. Barnes, Gary F. Egan, Combining images and anatomical knowledge to improve automated vein segmentation in MRI, NeuroImage, Volume 165, 15 January 2018, Pages 294-305, ISSN 1053-8119, https://doi.org/10.1016/j.neuroimage.2017.10.049.
Code to process these images and recreate them is available here: https://github.com/philgd/CVI-MRI
Details:
The space of all images is the 0.5mm MNI atlas.
Priors (SWI, QSM and VeinFrequency) denote the predictive power of the respective images throughout the brain. This was calculated using ten manually-traced vein maps. Predictive power of SWI and QSM were calculated by first fitting Gaussian Mixture Models to both.
VeinFrequencyMap is the combination of the co-registered manually-traced vein maps. The values sit between 0 and 1 (0-100%) representing the frequency with which each voxel was a vein in the manually-traced vein maps.
The training data, including SWI images, QSM images and manually traced vein masks are included (TrainingData). The transforms required to interpolate the images into MNI space are provided separately (Transforms).