This study deals with network-wide optimization of decentralized signal control based on Multi-Agent Reinforcement Learning (MARL) considering users’ route choices. We have proposed the decentralized signal control which theretically guarantees to yield network-wide optimum signal control. And, this paper extends the decentralized control to cases with users’ route choices. Validations of the proposed decentralized control are conducted on a simple two-intersection network through arterial signal control scenarios encompassing both undersaturated and oversaturated conditions as well as scenarios with/without route choice.