Abstract
Transformer has already proven its ability to model long-distance dependencies. However, medical images have strong local structures. Directly using Transformer to extract features would not only contain redundant information increasing the computational effort, but also be detrimental to extracting local details. Given these issues, we propose a network based on dynamic positioning and region-aware attention, which adopts a two-stage feature extraction strategy. In the shallow layer, we design Dynamic Positioning Attention (DPA). It will localize to the key feature information and construct a variable window for it, then perform attention calculation. DPA improves the learning ability of local details, reduces the amount of computation. At the deep level, Bi-Level Routing Attention (BRA) is used to discard irrelevant key–value pairs, achieve content-aware sparse attention for the deep dispersed semantic information, and improve computational efficiency. After several experiments, the results show that our method achieves advanced performance on different types of datasets.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.