Evaluation of the Influence of Data Sampling Schemes on Neural Diffusion Models

Abstract

The joint application of multiple diffusion models on single sampled dataset is becoming a hot topic in clinical research. This study investigated the influence of the three data sampling schemes on the quantification of neural diffusion models. The three sampling schemes compared were QGrid, Free and MDDW on the Siemens scanners. The diffusion models involved were diffusion tensor imaging (DTI), diffusion kurtosis imaging (DKI), neurite orientation dispersion and density imaging (NODDI) and mean apparent propagator (MAP) models. It was demonstrated that the results of NODDI and MAP were sensitive to the sampling schemes and the set of maximum b-value, while that of DTI and DKI were comparatively not sensitive to varying configurations. It was also shown that QGrid and Free schemes provided more consistent results. Thus the sampling scheme should be carefully selected in multi-center studies and studies with large sample size. QGrid and Free schemes are recommended for their advantages demonstrated in this study.

Keywords

• magnetic resonance imaging
• data sampling scheme
• diffusion model
• diffusion tensor imaging (dti)
• diffusion kurtosis imaging (dki)
• neurite orientation dispersion and density imaging (noddi)
• mean apparent propagator (map)