Abstract:Compressed Sensing MRI reconstructs images of the body's internal anatomy from undersampled measurements, thereby reducing the scan time - the time subjects need to remain still. Recently, deep neural networks have shown great potential for reconstructing high-fidelity images from highly undersampled measurements in the frequency space. However, one needs to train multiple models for different undersampling patterns and desired output image resolutions, since most networks operate on a fixed discretization. Such approaches are highly impractical in clinical settings, where undersampling patterns and image resolutions are frequently changed to accommodate different real-time imaging and diagnostic requirements. We propose a unified model robust to different measurement undersampling patterns and image resolutions in compressed sensing MRI. Our model is based on neural operators, a discretization-agnostic architecture. Neural operators are employed in both image and measurement space, which capture local and global image features for MRI reconstruction. Empirically, we achieve consistent performance across different undersampling rates and patterns, with an average 11 percent SSIM and 4dB PSNR improvement over a state-of-the-art CNN, End-to-End VarNet. For efficiency, our inference speed is also 1,400x faster than diffusion methods. The resolution-agnostic design also enhances zero-shot super-resolution and extended field of view in reconstructed images. Our unified model offers a versatile solution for MRI, adapting seamlessly to various measurement undersampling and imaging resolutions, making it highly effective for flexible and reliable clinical imaging. Our code is available at <a class="link-external link-https" href="https://armeet.ca/nomri" rel="external noopener nofollow">this https URL</a>.

A Unified Model for Compressed Sensing MRI Across Undersampling Patterns