Application of machine learning in optimizing b-value acquisition strategy of diffusion Magnetic Resonance Imaging

Abstract

The b-value acquisition strategy of diffusion Magnetic Resonance Imaging (dMRI) is very important for medical clinical application, especially the low b-value strategy. However, the choice of b-values is affected by several factors: for example, different tissue, different regions of tissue, the dependence of dMRI signals on b-values are different. Specifically, dMRI signals in areas with faster blood circulation may be more sensitive to low b-values (b<50 s/mm2); in addition, to obtain the diffusion or perfusion information from the diffusion-weighted (DW) signal, fitting methods are required, which also affected by low b-values. In this paper, Convolutional Neural Network (CNN), a machine learning based method is first used for learning the different characteristics of the DW signals in different regions of tissue and generated by different b-value acquisition strategy, and then analyse the dependence of DW signals on low b-values in different regions of the tissue. Finally, to study the dependence of the fitting methods on low b-values, which to determine the b-value acquisition strategy. The results show that the b-value acquisition strategy are different in different perfusion regions and using different fitting methods.