NODDInet: Diffusion parameter mapping using deep neural network trained by computer simulation data

A deep neural network, NODDInet, was developed to generate NODDI parameters (ICVF, ISOVF, OD, and FA) in 1 min. This network was trained using a computer simulation-generated training dataset only, and, therefore, is unbiased to experimental data and covers a wide range of the parameters. For the network input, the diffusion measurements of each shell were projected onto three 2D plains to reduce the input data size while preserving the geometric information of the diffusion measurements. The results demonstrate higher accuracy and faster processing time (x14) than a previous method (AMICO).