Abstract: Existing Convolutional Neural Network (CNN)-based auxiliary diagnosis methods fail to achieve ideal accuracy in the screening and staging tasks of pneumoconiosis. Therefore, this paper proposes a multi-branch structured model, namely MBM-Mamba, for the screening and staging of pneumoconiosis in chest X-ray images. MBM-Mamba introduces a novel six-directional scanning strategy under the SS2D framework, which explicitly captures two-dimensional local dependencies in linear time complexity through diagonal scanning. It then integrates prior information into the CNN to construct a detail enhancement module (Dem), thereby forming a local feature extraction module (CNN-Mamba) that significantly enhances the expression ability of subtle lesion information. Additionally, MBM-Mamba designs a global feature extraction module on the basis of the multi-head self-attention mechanism (W-MSA), effectively enhancing the ability to capture global context. MBM-Mamba integrates the two modules via multi-path residual connections, achieving synchronous hierarchical fusion of cross-structural features, thus enabling the model to better understand the overall distribution and fibrosis degree of pulmonary lesions. Validated on 1,760 real anonymized patient chest X-ray images, MBM-Mamba achieved an accuracy of 78.6% and an F1 score of 79%, both of which outperformed existing models.