This is the dataset *VerSe 2019* in **subject based data structure**. Image series of 141 patients (160 image series) are divided into a test (n=37), a validation (n=37), and a training set (n=67). Subject identifiers equal image series identifiers ('verseXXX'). For patient with 2 or 3 image series, new subject identifiers ('verse4XX') were introduced. This is an adaption of the Brain Imaging Data Structure (BIDS; https://bids.neuroimaging.io/). In VerSe2019, all scans have been reformatted to ASL orientation (i.e. origin of the xyz coordinate system anterior, superior, left) and down-sampled to a maximum of 1mm voxel length, i.e. 1mm x 1mm x 1mm maximum spatial resolution. Thicker slice thickness remained, i.e. no interpolation was performed. The centroids of the *subject based data structure* are in the image space (i.e. ASL) and refer to the image resolution, i.e. values are given in *voxels* (note that the challenge data format gives values in mm). This dataset can be used with it´s extension VerSe 2020 that is available in the same format here: https://osf.io/4skx2/ Please read the readme.txt and licence.txt: https://osf.io/pg6s3/ (Creative Commons Attribution-ShareAlike 2.0).