Hybrid CNN-MAMBA for Emphysema Classification

Hybrid CNN-MAMBA for emphysema classification. (a) The proposed framework for emphysema classification includes a low-level feature extractor $\mathcal{E}$, a hierarchical feature extractor $\mathcal{E}_f$, two Mamba Blocks (MB), and a classification head $\mathcal{C}$. (b) Mamba Block performing sequence modeling that captures long range dependencies in the feature representations ( $\tilde{Z}$ or $F$). (c) The in-domain self-supervised pretraining strategy facilitates learning meaningful feature representations from unlabeled CT scans, enhancing model robustness and generalization.