In this study, a two-stage post-earthquake recovery strategy optimization model is proposed for determining resource inputs and obtaining the optimal restoration sequence after earthquakes. An artificial neural network-based surrogate model for predicting the functionality of the interdependent transportation and healthcare system (ITHS) is established to avoid the significant time consumption arising from traffic assignment and path selection in functionality evaluation and recovery optimization. It is applied to the two-stage optimization and resilience assessment under different seismic scenarios, which can greatly improve computational efficiency. The resilience-based two-stage optimization decision-making method assisted by the surrogate model is applied to a real example and provides practical support for recovery strategies of the ITHS.