SpineMamba: Enhancing 3D spinal segmentation in clinical imaging through residual visual Mamba layers and shape priors

Zhiqing Zhang, Tianyong Liu, Guojia Fan, Na Li, Bin Li, Yao Pu, Qianjin Feng, Shoujun Zhou

Abstract: Highlights•Innovative design of a learnable 3D spinal shape prior module: This module, successfully integrated as a plug-and-play component within the network, learns spinal priors during training, refining the predicted spinal region boundaries and overcoming challenges related to inter-class similarity and intra-class variability, thereby significantly enhancing vertebrae segmentation performance.•Superior segmentation performance and robust generalization: The proposed method outperforms several state-of-the-art approaches across two independent datasets of different modalities, demonstrating strong generalizability and robustness, and providing valuable insights for the development of more efficient and effective SSM-based spinal segmentation methods.•The first Mamba-based 3D spinal segmentation model: By hybridizing SSM and CNN architectures, this model captures both local fine-grained features and long-range dependencies within images while maintaining linear computational complexity.