Vision Mamba and xLSTM-UNet for medical image segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning models like CNNs and Transformers are used for medical image segmentation.
• CNNs struggle with long-range dependencies due to limited receptive fields.
• Transformers capture global information but are computationally intensive for clinical use.

Purpose of the Study:

• Introduce VMAXL-UNet, a novel network combining SSM and xLSTM for efficient medical image segmentation.
• Address limitations of existing CNNs and Transformers in capturing both local and global features.
• Improve segmentation accuracy and robustness in complex medical imaging.

Main Methods:

• Developed VMAXL-UNet, integrating Visual State Space (VSS) and ViL modules in the encoder.
• VSS module uses SSM to capture long-range dependencies and extract features from distant regions.
• ViL module employs a gating mechanism for enhanced integration of local and global features.

Main Results:

• VMAXL-UNet demonstrated superior performance compared to traditional CNNs and Transformer-based models.
• The model excelled in capturing lesion boundaries and their distant correlations across multiple datasets (ISIC17, ISIC18, CVC-ClinicDB, Kvasir).
• Significant improvements in segmentation accuracy and robustness were observed.

Conclusions:

• VMAXL-UNet offers a promising approach for efficient and accurate medical image segmentation.
• The integration of SSM and xLSTM effectively balances local detail and global context.
• The model shows potential for practical application in clinical settings.