Deformable volume visualization is a cornerstone of modern medical simulations, such as virtual surgery and preoperative planning. However, sustaining high-resolution visual feedback in real-time remains a significant computational challenge due to the intensive resampling required during volume deformation. This study proposes a high-performance GPU-based framework designed to achieve interactive resampling speeds without compromising visual fidelity.

Our approach incorporates two synergistic acceleration strategies. First, a margin-refined block culling method is introduced to selectively process visually significant regions, effectively eliminating redundant computations while preserving boundary integrity near deformed interfaces. Second, we employ an optimized hardware resource management scheme that leverages shared memory tailored to 3D cell access patterns to reduce adjacency redundancy while balancing register pressure and memory throughput.

Across four datasets and two rendering modes, the framework accelerated the resampling stage by 2.30× to 6.58× relative to an already optimized baseline, while producing artifact-free reconstruction verified against full resampling. This acceleration keeps resampling interactive even for high-resolution volumes, removing it as the volume-resolution-bound bottleneck of the deformation pipeline.

The proposed framework provides a computational foundation for interactive deformable volume rendering. Its architecture is agnostic to the deformation algorithm, ensuring compatibility with physics-based models such as Position-Based Dynamics (PBD) and Chain-mail. By reducing the resampling cost of large-scale volume deformation, it supports the development of more responsive medical simulation systems; establishing clinical efficacy for any specific use remains a matter for dedicated validation.