DAMS-Mamba: integrating dynamic attention and multi-scale modelling for cerebral vessel segmentation

Recent developments in the fields of deep learning and computer vision have significantly impacted the segmentation of medical images, especially in the challenging task of cerebral vessel segmentation. The complexity of vascular structures and the need for high clinical accuracy present unique difficulties, particularly in segmenting small, thin blood vessels. This paper introduces DAMS-Mamba, a novel deep learning model designed to significantly improve the precision and efficiency of brain vessel segmentation. By leveraging advanced techniques such as multi-scale feature extraction, dynamic context-aware modelling and dynamic weighted feature fusion, DAMS-Mamba effectively utilises both local and global context information. This allows the model to capture the fine details of small vessels while maintaining the overall vessel structure, thus improving segmentation results in complex scenarios. The model was evaluated using publicly available datasets, including the Cerebral MRA Dataset Annotations and TubeTK public brain 3D vessel dataset, demonstrating an overall performance improvement of approximately 5% to 10% compared to existing methods. The experimental results underscore the potential of DAMS-Mamba for clinical applications, providing a powerful tool for vascular segmentation in the medical field. The study also suggests promising directions for future research, including optimising computational efficiency and extending the model to other medical image analysis tasks, further advancing the application of artificial intelligence in medical image analysis.