A computational approach has been developed for analysis 3D dislocation structures generated by continuum dislocation dynamics (CDD), with the goal of informing crystal plasticity (CP) models that are sensitive to dislocation microstructure. In CDD, the dislocation density on each slip system is represented by a vector field with a unique dislocation line direction at each point in space. The evolution of these density fields is governed by dislocation transport equations coupled with crystal mechanics. A streamline construction is proposed to obtain the characteristics of dislocation structures generated by CDD and estimate CP-relevant parameters from CDD simulations. The streamlines are obtained by travelling along the tangent of the dislocation density and velocity vector fields, enabling us to construct the dislocation lines and their motion trajectories in the crystal. As will be explained in the presentation, the streamlines are computed by solving a set of partial differential equations. Here, we use this approach, combined with statistical analysis to extract the microstructure parameters from CDD simulations that are relevant to crystal plasticity models accounting for the dislocation substructure. These parameters include the average mean free path and mobile dislocation segment length, as well as the dislocation wall volume fraction. The results show that, under monotonic loading, both the mobile dislocation segment length and dislocation mean free path decrease with strain, which is consistent with the models used in the literature. The results further show that the mobile dislocation segment length follows a log-normal distribution. Under cyclic loading, however, these parameters no longer vary monotonically and exhibit cyclic rise and fall that is influenced by the behaviour of the dislocation density. The overall approach has been lately turned into a general machine learning technique for extraction of continuum-relevant parameters from CDD simulations. This work has been performed in collaboration with Peng Lin (Beihang University, China), Vignesh Vivekanandan (formerly with Purdue University, USA), Khaled SharafEldin (Purdue University, USA), Gustavo Castelluccio (Cranfield University, UK), and Benjamin Anglin (Naval Nuclear Laboratory, USA).