Fig. 1.

Analysis flowchart. (STEP1) MR susceptibility-induced field, eddy current distortions, and motion artifacts of the individual diffusion weighted images were corrected, and diffusion tensor imaging metrics (FA, MD, AD, and RD map) were calculated by using the voxel-wise diffusion tensor modeling. (STEP2) After thresholding the John Hopkins University probabilistic white-matter pathway atlas, features were extracted by calculating the mean DTI metrics of the WM pathway regions. Using the elastic net logistic regression model, individuals were classified into either the preterm or full-term group. The most contributive features were then identified using parameters of the trained elastic net logistic regression model. (STEP3) Finally, partial correlation analysis was performed between the selected features and neurodevelopmental domains after removing the covariate effects with the multiple linear regression model. MR, magnetic resonance; FA, fractional anisotropy; MD, mean diffusivity; AD, axial diffusivity; RD, radial diffusivity; DTI, diffusion tensor imagin; WM, white matter;