Temperature field reconstruction in organ cryopreservation from sparse measurements using a physics-guided learning framework

Accurate prediction of transient three-dimensional temperature fields during organ cooling and rewarming is essential for maintaining viability in cryopreservation. However, this task is fundamentally constrained by the scarcity of experimentally measured data, the distribution discrepancy between simulated and real experimental data, and the strong nonlinearities associated with phase-transition processes. This paper proposes a Transient Temperature Query Network (TTQN) that integrates simulated and real experimental data by reformulating temperature-field reconstruction as a query-driven conditional regression task, and establishes a large-scale mixed dataset covering diverse animal tissues and organs. By incorporating multi-channel thermophysical parameters, multi-scale feature-extraction modules, and physics-guided loss functions, the proposed model enables precise temperature prediction at arbitrary internal positions of organs under sparse observation conditions. The proposed TTQN demonstrates superior prediction performance compared with several benchmark models, achieving stable reconstruction accuracy under limited observation conditions and significantly reducing simulation-to-experiment discrepancies. This study establishes a scalable and generalizable paradigm for three-dimensional temperature-field prediction in biomaterials, providing core technical support for precise thermal-regulation strategies during organ preservation.