Table 1.

Overview of 2D- and 4D-Flow deep learning-based reconstructions.

Manuscript	Imaging	Network	Training	Undersampling	Recon Reduction
Vishnevskiy [97]	4D-Flow	Unrolled	n = 11	12.4 ≤ R ≤ 13.8	30×
Haji-Valizadeh [98]	2D-Flow	3D U-Net	n = 510	R = 28.8	4.6×
Cole [99]	2D-Flow	2D U-Net	n = 180	R ≤ 6	N/A
Jaubert [100,101]	2D-Flow	3D U-Net	n = 520	R = 18	15×
Oscanoa [39]	2D-Flow	2D U-Net	n = 155	R = 8	N/A
Kim [102]	4D-Flow	Unrolled	n = 140	R ≤ 6	N/A
Nath [103]	4D-Flow	2D U-Net	n = 18	R = 2.5, 3.3, 5	7×

Reconstruction time improvement is relative to compressed sensing (CS).