Focal brain cooling (FBC) is a treatment for refractory epilepsy to suppress epileptic discharges from an epileptic focus. Our wearable FBC system under development consists of a recirculating coolant apparatus and a battery located extracorporeally, and a cooling device made of titanium with water channels inside which is embedded in the skull. An optimal channel design of the cooling device is needed to cool the brain efficiently. Although clinical and animal studies are required to design the cooling device, the number of experiments should be reduced, and finite element (FE) simulation be utilized instead. However, trial-and-error simulations impose a heavy computational burden. One option to deal with this problem is to use surrogate modeling, which is a data-driven approach that mimics the behavior of simulation models with a low computational burden. We optimized the channel structure by combining the surrogate model approach and non-dominated sorting genetic algorithm II (NSGA-II), which is a multi-objective optimization algorithm. The objective was to design a cooling device that simultaneously achieves high cooling performance and low pressure drop since low pressure loss contributes to miniaturization and energy saving of the recirculating apparatus. The optimization result showed that pressure loss was reduced by 28.7% in comparison with the design variables determined through experiments without deteriorating the cooling performance.