In the cooling systems of a nuclear fusion reactor (NFR), anomaly events may occur where the flow of coolant unexpectedly decreases, such as in a loss-of-flow accident (LOFA). From the viewpoint of reactor safety design, ITER should be capable of handling subcooled boiling in coolant duct lines due to a sudden LOFA under severe operating conditions where the reactor’s divertor is under a steady-state massive heat load of 10 . To detect anomaly events, we performed a proof-of-concept (PoC) experiment and propose a novel method of state sensing of the coolant duct line based on sound emitted from boiling. In the PoC pool boiling experiment, using a hydrophone, we acquired sound from water boiling where the heat transfer surface was made of a divertor alloy [copper-chromium-zirconium (CuCrZr)] or copper (Cu) as the reference material. Notably, microbubble emission boiling (MEB) occurred at the highest heat flux in the case of CuCrZr alloy as well as Cu. This indicates that MEB can also occur in an actual NFR. In this study, we developed deep neural network (DNN) models to identify boiling states from acquired sound. Our well-trained DNN models could distinguish between the boiling sound of Cu and the CuCrZr alloy of the heat transfer surface material. The regression DNN models achieved a coefficient of determination of approximately 99%. Based on the PoC study, we have demonstrated that DNN models of boiling acoustics have the potential to identify boiling states inside coolant duct lines in the NFR, where the heat flux ranges from the low region to the high region with MEB occurrence.