Data for Publication FlowVN Trained on a Single Dataset Enables Rapid Reconstruction of Highly Accelerated 4D Flow MRI Across Multiple Sites

This dataset contains training and test inputs for the refactored FlowVN pipeline for highly accelerated 4D Flow MRI reconstruction.

The training set consists of HDF5 files containing complex-valued image data and coil sensitivity data stored as separate real and imaginary components. In each training file, the image tensors are organized by velocity encodings, cardiac time frames, slices, and in-plane image dimensions, while the coil tensors are organized by coil channels, slices, and in-plane image dimensions. In this dataset, the image data correspond to 4 velocity encodings, 25 time frames, 19 slices, and image dimensions of 112 x 112 pixels. The coil data correspond to 5 coil channels over the same 19 slices and 112 x 112 image grid.

The test set consists of MATLAB files containing reconstructed image data, undersampled k-space data, and coil sensitivity maps for one test volume across multiple acceleration factors. The image data are organized as 112 x 112 in-plane image dimensions, 19 slices, 25 time frames, and 4 velocity encodings. The k-space data follow the same structure with an additional coil-channel dimension of 5 coils. Coil sensitivity maps are provided for the same 112 x 112 image dimensions and 19 slices across 5 coil channels. A corresponding sampling mask is also included for the test volume.