Multi-contrast MRI Super-Resolution via a Multi-stage Integration Network

Super-resolution (SR) plays a crucial role in improving the image quality of magnetic resonance imaging (MRI). MRI produces multi-contrast images and can provide clear display soft tissues. However, current super-resolution methods only employ single contrast, or use simple fusion mechanism, ignoring rich relations among different contrasts, which are valuable for SR. In this work, we propose multi-stage integration network (i.e., MINet) SR, explicitly models dependencies between at stages to guide particular, our MINet first learns hierarchical feature representation from multiple convolutional each different-contrast image. Subsequently, introduce module mine comprehensive representations images. Specifically, matches with all other features, integrated terms their similarities obtain an enriched representation. Extensive experiments on fastMRI real-world clinical datasets demonstrate that 1) outperforms state-of-the-art SR various metrics 2) is able excavate complex interactions features stages, leading improved target-image quality.