SANGRE: a Shallow Attention Network Guided by Resolution Expansion for MR Image Segmentation

Magnetic Resonance (MR) imaging plays a vital role in clinical diagnostics and treatment planning, with the accurate segmentation of MR images being of paramount importance. Vision transformers have demonstrated remarkable success in medical image segmentation; however, they fall short in capturing the local context. While images of larger sizes provide broad contextual information, such as shape and texture, training deep learning models on such large images demands additional computational resources. To overcome these challenges, we introduce a shallow attention feature aggregation (SAFA) module to progressively enhance features’ local context and filter out redundant features. Moreover, we use feature interactions in a resolution expansion guidance (REG) module to leverage the wide contextual information from the images at higher resolution, ensuring adequate exploitation of small class features, leading to a more accurate segmentation without a significant increase in FLOPs. The model is evaluated on two dynamic MR datasets for speech and cardiac cases. The proposed model outperforms other state-of-the-art methods. The codes are available at https://github.com/Yhe9718/SANGRE .