Projection Domain

Projection domain methods are transforming various image processing and machine learning tasks by operating directly on the raw projection data rather than on reconstructed images. Current research focuses on developing novel neural network architectures, including diffusion models and iterative networks, to improve tasks like 3D reconstruction from single images, low-dose CT/SPECT imaging, and semantic segmentation. These advancements aim to enhance image quality, reduce computational costs, and improve the robustness of algorithms, particularly in applications with limited data or noisy projections. The resulting improvements have significant implications for medical imaging, computer vision, and other fields relying on efficient and accurate image analysis.