Data Driven Reconstruction

Data-driven reconstruction uses machine learning to improve image reconstruction from incomplete or noisy measurements across various imaging modalities, aiming for higher quality and computational efficiency. Current research focuses on developing and analyzing novel neural network architectures, such as invertible residual networks and unrolled iterative methods, often incorporating physics-based priors to enhance accuracy and generalization across diverse datasets and measurement conditions. These advancements are crucial for improving the speed and quality of medical imaging, radio astronomy, and other applications where high-fidelity reconstruction from limited data is essential.