The paper introduces a lightweight 4D Gaussian Splatting (LGS) framework for efficient surgical scene reconstruction. LGS addresses the challenges of high-dimensional Gaussian attributes and high-resolution deformation fields, which are common issues in 4D Gaussian Splatting (4D-GS) techniques. To improve efficiency, LGS employs three main strategies: Deformation-Aware Pruning (DAP), Gaussian-Attribute Pruning (GAP), and Feature Field Condensation (FFC). DAP identifies and prunes Gaussians that are not crucial for deformation, GAP reduces the dimensionality of Gaussian attributes by simplifying textures and lighting in non-critical areas, and FFC compactly represents the high-resolution spatial-temporal feature field. Experiments on public datasets show that LGS achieves a compression rate of over 9× while maintaining high visual quality and real-time rendering efficiency, making it suitable for practical deployment in robotic surgical equipment. The project page is available at <https://lgs-endo.github.io/>.The paper introduces a lightweight 4D Gaussian Splatting (LGS) framework for efficient surgical scene reconstruction. LGS addresses the challenges of high-dimensional Gaussian attributes and high-resolution deformation fields, which are common issues in 4D Gaussian Splatting (4D-GS) techniques. To improve efficiency, LGS employs three main strategies: Deformation-Aware Pruning (DAP), Gaussian-Attribute Pruning (GAP), and Feature Field Condensation (FFC). DAP identifies and prunes Gaussians that are not crucial for deformation, GAP reduces the dimensionality of Gaussian attributes by simplifying textures and lighting in non-critical areas, and FFC compactly represents the high-resolution spatial-temporal feature field. Experiments on public datasets show that LGS achieves a compression rate of over 9× while maintaining high visual quality and real-time rendering efficiency, making it suitable for practical deployment in robotic surgical equipment. The project page is available at <https://lgs-endo.github.io/>.