Zero-Shot Novel View Synthesis of Wrist X-Rays Using Latent Diffusion Model

X-ray imaging plays a crucial role in diagnosing and monitoring injuries such as distal radius fractures, which are among the most common musculoskeletal injuries. However, challenges such as patient comfort, radiation exposure, and cost/time constraints make it difficult to obtain a large number of X-ray views. We present a view-conditioned latent diffusion model capable of synthesizing new X-ray views of the wrist from a single X-ray input, enhancing the diagnostic capabilities and clinical utility of X-ray imaging. Preliminary results demonstrate the model’s capability to generate realistic and clinically relevant X-ray views of the wrist from a single input, showing strong zero-shot performance on new patient anatomy and true radiographs.