Medical Image Super-Resolution
Overview
Medical image super-resolution aims to recover sharper, often more isotropic images from scans with thick slices, low signal, or limited acquisition data, across modalities such as MRI, CT, PET, and histopathology. Methods include supervised CNNs, GANs, transformers, and diffusion models, as well as self-supervised and implicit neural representations that fit a single subject; physics-based reconstruction and anatomical or cross-modal guidance help constrain the added detail. A central challenge is that models trained on synthetic downsampling may not generalize to real scanner degradations, and visually convincing detail can be hallucinated or erase small lesions. Evaluation is therefore expanding beyond PSNR and SSIM to data consistency, uncertainty, anatomical and lesion preservation, and downstream tasks, while the amount of subject-specific detail recoverable from the original scan remains a fundamental limit.
Latest papers
57 papers8 papers in the last four weeks, up 33% on the four weeks before. 0.0% of all new papers.