Render-FM: Feedforward Model for Real-time Photorealistic Volumetric Rendering

Render-FM is a novel feedforward model for photorealistic volumetric rendering of CT scans, achieving a 500x speedup by regressing 6D Gaussian Splatting (6DGS) parameters in just 2.8 seconds per scan, compared to hours for traditional methods like NeRF. It incorporates Anatomy-Guided Priming (AGP) to leverage segmentation masks and transfer functions, enhancing its ability to generalize across different anatomies and support real-time rendering without extensive preparation. This advancement facilitates clinical workflows by providing immediate, high-quality visualizations, significantly improving the efficiency of medical imaging applications.