Design and application of ISSA-BP neural network model for predicting soft tissue relaxation force

Purpose: Accurate biomechanical modeling is crucial for enhancing the realism of virtual surgical training. This study addressed the computational cost and complexity associated with traditional viscoelastic models by incorporating neural network algorithms, thereby augmenting the predictive capability of soft tissue modeling. Methods: To address these challenges, the present study proposed a novel biomechanical modeling approach. The approach establishes a relaxation prediction model based on the backpropagation (BP) neural network and optimizes it using an enhanced sparrow search algorithm (ISSA). This hybrid method leverages the dynamic characteristics of forceps to predict the relaxation force of soft tissues more accurately. The ISSA optimizes the model by integrating chaos mapping, nonlinear inertia weight, and vertical–horizontal crossover strategy, which helps overcome the issue of local optima and boosts the predictive performance. Results: The experimental results demonstrated that the R₂ values reached 0.9956 for the pig kidney and 0.9896 for the pig stomach, indicating the model’s exceptional precision in predicting relaxation forces. Conclusions: The relaxation force prediction model based on ISSA-BP neural network provides excellent predictive performance, offering a new and effective strategy for biomechanical modeling of soft tissues in virtual surgical systems.