Abstract Introduction: Breast cancer is a complex disease with no simple solution, which is why researchers are tackling it from every angle. A relatively new effort of predictive oncology is to develop a systematic method for predicting the future status of an individual breast tumor given representation of the initial conditions and an appropriate mathematical model. Using a medical imaging-based approach, the goal of this preclinical research project was to obtain high resolution images of the tumor angiogenic network and predict cancer growth using a novel mathematical model that we recently develop. Methodology: In the proposed model, we consider each cell and vessel as an agent and decisions are made based on the state-action-reward-state-action (SARSA) concept, which is an incremental reinforcement learning algorithm. Two types of cells (i.e. cancer and normal) and vessels (i.e. tip and stalk) are used in the environment. This spatial collection of cells and vessels then have six (i.e. apoptosis, hypoxia, necrosis, migration, proliferation and quiescence) and three (i.e. branch, expansion and sprout) allowable actions. The model agent has a high degree of autonomy and performs actions based on the knowledge it receives from the surrounding environment. One of the more important parts of our model is the release and diffusion of select nutrients into the tumor environment, e.g. oxygen, vascular endothelial growth factor (VEGF), etc. Using a breast cancer xenograft mouse model, animals were administrated an intravascular contrast agent (200 microliters, ExiTron Nano 12000, Miltenyi Biotec) and tumors were scanned using an ultrahigh resolution computed tomography (CT) system (OI/CT, MILabs). The tumor microvascular network was segmented and used as an input and initial state of our model that then simulated cancer growth. Subsequent CT scans (on the order of weeks) were performed to validate model accuracy. Results: Initial results demonstrate that the SARSA model can simulate tumor growth. As the concentration of available nutrients around the tumor decreases, normal cells began to decline sharply and were eliminated by selecting the apoptotic phenotype. Cancer cells were more resistant to nutrient deficiencies during simulation. Consequently, these cancer cells proliferated, and tumor volume increased. Our findings also revealed that decreased oxygen availability (i.e. hypoxia) stimulated VEGF production and growth of the microvascular network, which agreed with repeat in vivo CT imaging results. Conclusion: A predictive model of breast cancer growth was developed and preliminary preclinical results using in vivo medical images of breast cancer-bearing animals highlighted the utility of this new oncological tool. Citation Format: M. Hossein Zangooei, Ryan Margolis, Kenneeth Hoyt. A new computational model for visualization and prediction of breast cancer growth based on reinforcement learning [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS3-22.