CoReFuse-Med
A fusion framework for multimodal medical image segmentation that employs a two-stage design of shallow-layer decontamination and deep-layer bias correction to address fusion failure caused by differences in sampling quality across modalities.
Timeline (last 90 days)
CoReFuse-Med was proposed by Central China Normal University together with Huazhong University of Science and Technology, Guangzhou University of Chinese Medicine and the University of North Carolina at Chapel Hill.
The paper "When Fusion Fails: Corruption-Aware Rebalanced Fusion for Multi-Modal Medical Image Segmentation" was accepted by ACM MM 2026.
The authors limit the problem scope to quality degradation caused by resolution differences, and do not address cases of misaligned modalities or completely missing modalities.
CoReFuse-Med splits fusion into two stages, keeping an overall three-stream, symmetric U-shaped segmentation network structure that contains three modules: LACE, IMSH and MBCF.
On the EPVS dataset, CoReFuse-Med achieves a DSC of 0.7326 and an HD95 of 7.35, outperforming MedNeXt's 0.7085 and 8.17.
On the WMH dataset, CoReFuse-Med achieves a DSC of 0.80, an HD95 of 6.22 and an IAVD of 0.171, with only 2.665M parameters.
All Facts (6)
CoReFuse-Med was proposed by Central China Normal University together with Huazhong University of Science and Technology, Guangzhou University of Chinese Medicine and the University of North Carolina at Chapel Hill.
50%UnverifiedThe authors limit the problem scope to quality degradation caused by resolution differences, and do not address cases of misaligned modalities or completely missing modalities.
50%UnverifiedCoReFuse-Med splits fusion into two stages, keeping an overall three-stream, symmetric U-shaped segmentation network structure that contains three modules: LACE, IMSH and MBCF.
50%UnverifiedOn the EPVS dataset, CoReFuse-Med achieves a DSC of 0.7326 and an HD95 of 7.35, outperforming MedNeXt's 0.7085 and 8.17.
50%UnverifiedOn the WMH dataset, CoReFuse-Med achieves a DSC of 0.80, an HD95 of 6.22 and an IAVD of 0.171, with only 2.665M parameters.
50%UnverifiedThe paper "When Fusion Fails: Corruption-Aware Rebalanced Fusion for Multi-Modal Medical Image Segmentation" was accepted by ACM MM 2026.
50%