Dual-Branch Encoder Network with SAM Prompts Improves Brain MRI Lesion Segmentation

Brain MRI lesion segmentation with DBENet dual-branch network achieves Dice 0.8621 on ISLES 2022 and 0.8261 on BraTS 2018, outperforming state-of-the-art methods.

• DBENet uses dual-branch spatial-frequency encoder with SAM prompts and multi-scale attention fusion for brain MRI lesion segmentation.
• In experiments on ISLES 2022 and BraTS 2018, DBENet achieved Dice scores of 0.8621 and 0.8261, outperforming prior methods.
• Ablation studies confirmed the contributions of spatial-frequency fusion and multi-scale attention modules.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.