The paper presents the results of a series of combined mechanical and in-situ morphological investigations on highstrength strain-hardening cement-based composites (SHCC). Tension and compression experiments were performed in a CT scanner employing a dedicated mechanical testing rig. The in-situ microtomographic scans enabled correlating the measured specimen response with relevant microstructural features and fracture processes. The microstructural segmentation of SHCC was performed in the framework of Deep Learning and it targeted an accurate segmentation of pores, fibers and aggregates. Besides their accurate volumetric representation, these phases were quantified in terms of content, size and orientation. The fracture processes were monitored at different loading stages and Digital Volume Correlation (DVC) was employed to spatially map the strains and cracks in the specimens loaded in compression. The DVC analysis highlighted the effect of loading conditions, specimen geometry and material heterogeneity at the mesolevel on the strain distribution and fracture localization.