Background and objectiveSoft-tissue dynamics plays an essential role in the mechanical functions of the human body. Numerical approaches using finite element modeling and mass-spring framework have been currently used to estimate the biological soft tissue dynamics. However, these approaches still have important computational cost due to the use of a mesh configuration in the formulation of the dynamic equilibrium equation and unstable convergence issue. MethodsWe present in this study a novel approach based on the deep learning framework to predict the deformation of soft-tissues. In particular, the Long Short-term Memory (LSTM) neural network and deep neural network (DNN) were used to deal with high-frequency oscillation signals. Different learning strategies (with and without data dimension reduction) were also applied. A simulation-based database was generated using our HyperMSM model for training and testing purposes. The application of the proposed approach on the childbirth simulation was addressed. ResultsUsing the root mean square error, the LSTM- and DNN-derived deformation deviation range from 0.139 mm to 1.062 mm (0.266%–2.028%) for both training and testing processes. The Pearson correlation coefficient of 0.994–0.999 demonstrates the strong similarity between predicted outputs and ground truth data. ConclusionsThis present work showed the capacity of the deep learning neural networks to predict complex physiological signals of the human body functions. As a perspective, this approach will be coupled with the Hololens device toward a novel interactive childbirth training tool with real-time feedback on the soft tissue deformations.