Fig. 1: Digital twin overview.

A Describes the format of the digital twin which consists of both physical and digital components. The physical component informs the status of its digital counterpart based on inputs from observational data and physically delivered controls (e.g., chemotherapy). The digital state is calibrated based on the physical components, then returns quantities of interest to derive or optimize associated rewards. The rewards can then be used as feedback to inform the next set of controls in the physical space. We provide personalized regimens that theoretically provide improved response based on simulation results. B Provides an overview of the clinical and modeling timeline. The physical space is represented by the blue circles (imaging) and orange rectangles (A/C treatment). The first two visits (V1 and V2) are used to calibrate the digital state, which can predict rewards at the end of the simulation. The rewards are optimized (e.g., minimized in the problems described) by adjusting the treatment schedule for the third and fourth cycles, consisting of one cycle of unknown variables (purple rectangles) which are delivered twice, sequentially.