The OCMR dataset is comprised of HDF5 files, with data in each file following the ISMRMRD format. The entire dataset (ocmr_cine.tar.gz) can be downloaded by clicking here. The GitHub repository, https://github.com/MRIOSU/OCMR, provides step-by-step instructions to download and read the data in Matlab and Python. Note, the Python option eliminates the need to explicitly download the large tar file and allows downloading a subset of the dataset based on data file attributes.
Once a data file is read into Matlab or Python, it yields the k-space array, kData, and a structure, param. The kData array has nine dimensions: [kx, ky, kz, coil, phase, set, slice, rep, avg], which represent frequency encoding, first phase encoding, second phase encoding, coil, phase (time), set (velocity encoding), slice, repetition, and number of averages, respectively. For example, a cine stack with frequency encoding size 160, phase encoding size 120, number of coils 18, number of frames 60, number of slices ten will generate kData with these dimensions: 160x120x1x18x60x1x10x1x1. The second output, param, provides pertinent acquisition parameters. For example, param.FOV, param.TRes, param.flipAngle_deg, param.sequence_type specify field-of-view, temporal resolution, flip angle, and the type of sequence, respectively. More formal documentation of OCMR can be found in this arXiv article.