IVIM reconstruction from highly under-sampled DW-PROPELLER acquisition data via synthetic data-driven physics-informed deep learning

A synthetic data-driven physics-informed network (SDDPI-Net) was proposed for intravoxel incoherent motion (IVIM) mapping based on highly under-sampled diffusion-weighted turbo spin echo PROPELLER (DW-TSE-PROPELLER) data. This reconstruction network directly estimated distortion-free and artifacts-free IVIM parameters by explored data redundancy in the k-b space and IVIM bi-exponential model with synthetic training data. The results of human brain experiments show that our method can significantly improve the accuracy of IVIM maps with 6´ under-sampled DW-TSE-PROPELLER than other methods.