Differentiable Ultrasound Simulation for Optimization Challenges
This paper introduces a fully differentiable end-to-end ultrasound simulation framework based on full-path Monte Carlo ray tracing. It extends existing ray-based methods by propagating gradients from image-space losses back through the entire simulation pipeline, enabling gradient-based optimization for scene and acquisition parameters. This advancement addresses the limitations of high-fidelity, but computationally expensive, full-wave solvers, offering a practical solution for tasks like calibration and inverse parameter estimation.