Utilizing Q-Learning to Generate 3D Vascular Networks for Bioprinting Bone

Abstract

ABSTRACTBioprinting is an emerging tissue engineering method used to generate cell-laden scaffolds with high spatial resolution. Bioprinted vascularized bone grafts are a potential application of this technology that would meet a critical clinical need, since current approaches to volumetric bone repair have significant limitations. However, generation of vascular networks suitable for bioprinting is challenging. Here, we propose a novel Q-learning approach to quickly generate 3D vascular networks within patient-specific bone geometry that are optimized for bioprinting. First, the inlet and outlet locations are specified and the scenario is modeled using a grid world for initial agent training. Next, the path planned in the grid world environment is converted to a Bezier curve, which is then used to generate the final 3D vascularized bone model. The vessels generated using this procedure have minimal tortuosity, which increases the likelihood of successful bioprinting. Furthermore, the ability to specify inlet and outlet position is necessary for both surgical feasibility as well as generation of more complex vascular networks. In total, this study demonstrates the reliability of our reinforcement learning method for automated generation of 3D vascular networks within patient-specific geometry that can be used for bioprinting vascularized bone grafts.