SliceMamba for Medical Picture Segmentation
Authors: Chao Fan, Hongyuan Yu, Luo Wang, Yan Huang, Liang Wang, Xibin Jia
Summary: Regardless of the progress made in Mamba-based medical picture segmentation fashions, present strategies using unidirectional or multi-directional characteristic scanning mechanisms fail to effectively mannequin dependencies between neighboring positions within the picture, hindering the efficient modeling of native options. Nevertheless, native options are essential for medical picture segmentation as they supply very important details about lesions and tissue constructions. To handle this limitation, we suggest a easy but efficient methodology named SliceMamba, a domestically delicate pure Mamba medical picture segmentation mannequin. The proposed SliceMamba consists of an efffcient Bidirectional Slice Scan module (BSS), which performs bidirectional characteristic segmentation whereas using diversified scanning mechanisms for distinct options. This ensures that spatially adjoining options keep proximity within the scanning sequence, thereby enhancing segmentation efficiency. Intensive experiments on pores and skin lesion and polyp segmentation datasets validate the effectiveness of our methodology. △ Much less
