基于多粒度聚合与高频增强的DSA脑血管分割OA

Digital Subtraction Angiography(DSA)is the"gold standard"to diagnose cerebrovascular diseases.It can help doc-tors diagnose and treat these diseases.However,although existing methods have made certain progress in the field of medical im-age segmentation,there are still problems such as the loss of key area information and insufficient attention to small blood ves-sels.This paper proposes a Multi-Granularity Aggregation and High-Frequency Enhancement Network(MHF-Net).This net-work includes a multi-granularity aggregation module and a high-frequency enhancement module.These modules improve the ac-curacy and robustness of segmentation.The multi-granularity aggregation module in the encoder uses three complementary convo-lution branches to enhance features.The standard convolution branch can maintain the spatial coherence between blood vessels and the background,the variance modulation branch can capture global context information,and the depth convolution branch can enhance the features of boundaries and textures.In the skip connection layer,high-frequency features are integrated by us-ing channel and spatial attention to guide the model to focus on small blood vessels and blurred boundary regions.On the DIAS and DSCA datasets,this paper conducts comparative experiments using eight methods.The results show that MHF-Net can well preserve the information of key regions and enhance the attention to small blood vessels,achieving better blood vessel segmenta-tion results.This architecture can effectively address the problems of key region information loss and insufficient attention to small blood vessels in existing methods,demonstrating higher accuracy and robustness in the DSA cerebral blood vessel segmen-tation task.