Parameter Identification Method for Thromboelastography Detection Module

The thromboelastography (TEG) analyzer is an internationally recognized instrument used for intraoperative diagnosis of coagulopathy. For clinicians, employing TEG enables accurate assessment of thrombotic conditions during major surgical procedures, thereby enhancing surgical success rates. The thromboelastography detection module is the most critical component of the TEG analyzer, where identifying whether its system parameters are functioning correctly holds significant diagnostic value for evaluating coagulation function and surgical risk in clinical medicine. In this study, we proposed an LSTM-PINN model for the task of parameter identification described above. We utilized the angular displacement response signal of the thromboelastography detection module as the input. The traditional Physics-Informed Neural Networks (PINNs) model was selected as a comparative benchmark. We explored the selection of hyperparameter for the loss function and compared the model’s capabilities in signal and parameter identification. Ultimately, we determined the optimal hyperparameter suited for this parameter identification task. Furthermore, the LSTM-PINN model demonstrated superior signal identification capability and parameter recognition performance, exhibiting greater stability than the PINN model.