Versatility of OSAR across Medical Imaging Modalities

Various forms of artifacts with diverse patterns and combinations can be observed in medical images. These artifacts vary depending on the effectiveness of deep learning-based artifact reduction techniques in medical imaging is determined by factors such as scan setup, equipment quality, patient characteristics, and environmental conditions. However, these methods are limited by their reliance on training data that only covers predefined artifact patterns, making them impractical in clinical settings. This study introduces One-Shot Medical Image Artifact Reduction (OSAR), which seeks to address these limitations. A deep learning approach that maximizes its potential without the need for pre- trained general networks. Specifically, OSAR utilizes synthesized data from the input image during testing to train a lightweight, image-specific artifact removal network. This enables OSAR to effectively address a wide range of additive artifacts present in medical images that may not be represented in existing large training datasets.