# LBMNet: a hybrid multi-scale CNN–Mamba framework for enhanced 3D stroke lesion segmentation in MRI

## TL;DR

This paper introduces LBMNet, a new framework combining CNN and Mamba models to improve MRI-based stroke lesion segmentation, especially for small lesions.

## Contribution

LBMNet introduces a hybrid CNN–Mamba architecture with novel modules for multi-scale encoding and efficient global context modeling.

## Key findings

- LBMNet achieved a Dice coefficient of 67.57% on ATLAS v2.0 and 82.03% on ISLES 2022.
- The model outperforms existing CNN, Transformer, and hybrid models in small lesion segmentation.
- BSC-Mamba and BAGF modules enhance local and global feature modeling efficiently.

## Abstract

Brain stroke is one of the leading causes of death and disability worldwide, and accurate lesion segmentation from MRI is critical for clinical diagnosis and treatment planning. However, existing methods struggle with the high variability of stroke lesions in size and morphology. In particular, they fail to detect small lesions due to the limited receptive fields of CNNs and the computational inefficiency of Transformer-based approaches. To address these challenges, we propose LBMNet, a novel CNN–Mamba network that integrates multi-scale convolutional encoding with Mamba-based decoding.

Owing to the high heterogeneity of stroke lesions, the encoder design employs a top-down LSC module to capture cross-scale representations. The decoder designs the BSC-Mamba (Bidirectional Spatial Context Mamba) model, integrating bidirectional state space modeling with adaptive spatial convolutions to enhance local feature information while modeling global dependencies with linear complexity. Furthermore, asymmetric adaptive gated feature fusion (BAGF) bridges the semantic gap by selectively merging encoder and decoder features, suppressing redundant information whilst highlighting critical lesion details.

Extensive experiments on two benchmark datasets demonstrate state-of-the-art performance, achieving Dice coefficients of 67.57% on ATLAS v2.0 and 82.03% on ISLES 2022. Compared with existing CNN, Transformer, and hybrid models, LBMNet shows significant improvements in small lesion segmentation. This study presents a robust and efficient framework with strong clinical potential for accurate stroke lesion segmentation across diverse lesion sizes and morphologies.