MicFormer⁃HMD：多分支融合与混合门控驱动的心脏图像分割方法

doi:10.12422/j.issn.1672-6952.2026.03.012

摘要/Abstract

摘要：

医学图像分割是计算机视觉领域的关键技术，尤其是在处理计算机断层扫描（CT）和心脏磁共振成像（MRI）多模态医学影像时能提供关键诊断信息，但现有技术在模态协同建模、结构边界精准表达以及多尺度语义信息的有效整合上仍存在明显不足。为应对这些挑战，提出了改进传统MicFormer架构的MicFormer?HMD。设计了混合门控模块，通过参数化卷积与门控函数实现跨模态交互前的动态特征选择，具备自适应抑制噪声的优势并能增强判别性特征表达；设计了多分支特征融合模块，采用多分支的空洞卷积结构和双重注意力校准机制，提升了模型对多尺度上下文信息的捕捉和融合能力；采用了动态蛇形卷积，其可变形卷积核能自适应贴合心脏解剖结构的复杂形态，增强了几何感知能力。该MicFormer?HMD架构在心脏图像分割任务中展现出显著优势，特别是在保持薄壁组织连续性和复杂血管连通性方面表现突出。

关键词: 医学图像分割, 计算机视觉, 多模态, 卷积, 注意力机制

Abstract:

Medical image segmentation serves as a pivotal technology in computer vision, particularly capable in providing critical diagnostic information when processing multi?modal medical images like CT and MRI. However, existing techniques still exhibit significant limitations in modality collaborative modeling, precise structural boundary representation, and effective integration of multi?scale semantic information. To address these challenges, this paper proposes MicFormer?HMD, an enhanced architecture that improves upon the traditional MicFormer framework.A Hybrid Gating Module is designed to achieve dynamic feature selection before Cross?Modal interaction through parameterized convolutions and gating functions, enabling adaptive noise suppression and enhances discriminative feature representation. Then, we develop a Multi?Branch Fusion Attention module that employs a Multi?Branch dilated convolution architecture and a dual attention calibration mechanism,significantly improving the model's capability in capturing and integrating multi?scale contextual information. Dynamic Snake Convolution is incorporated, whose deformable kernels adaptively conform to the complex morphology of cardiac anatomical structures, thereby strengthening geometric perception. The proposed MicFormer?HMD architecture demonstrates remarkable advantages in cardiac image segmentation tasks, showing particular improvements in maintaining thin?walled tissue continuity and complex vascular connectivity.

Key words: Medical image segmentation, Computer vision, Multi?modal, Convolution, Attention mechanism

中图分类号:

TP18

崔硕, 郭来德, 李博文, 窦玲, 曾慧, 李腾, 雷昭鹏. MicFormer⁃HMD：多分支融合与混合门控驱动的心脏图像分割方法[J]. 辽宁石油化工大学学报, 2026, 46(3): 90-98.

Shuo CUI, Laide GUO, Bowen LI, Ling DOU, Hui ZENG, Teng LI, Zhaopeng LEI. MicFormer⁃HMD: A Multi⁃Branch Fusion and Hybrid⁃Gating Driven Approach for Cardiac Image Segmentation[J]. Journal of Liaoning Petrochemical University, 2026, 46(3): 90-98.

图/表 10

图1 学习率对模型性能的影响（a） Dice相似系数（b） 95%豪斯多夫距离（c）平均对称表面距离

Fig.1 Effect of learning rates on model performance

图2 损失权重对模型性能的影响（a） Dice相似系数（b） 95%豪斯多夫距离（c）平均对称表面距离

Fig.2 Effect of loss weights on model performance

图3 不同模型的心脏边界平滑性对比

Fig.3 Comparison in the smoothness of cardiac boundaries among different models

图4 不同模型的心房结构完整性对比

Fig.4 Comparison of the structural integrity of atrial structures among different models

图5 不同模型的预测洁净度对比

Fig.5 Comparison of predicted cleanliness among different models

表1 各模型在MMWHS数据集上的性能对比

Table 1 Performance comparison of different models on the MMWHS dataset

模型	Dice相似系数/%	95%豪斯多夫距离/mm	平均对称表面距离/mm
U⁃Net	80.49	14.64	4.40
Swin⁃UNet	81.32	12.76	3.80
R50⁃AttUNet	82.74	11.83	3.60
TransBTS	83.92	11.25	3.44
MicFormer	85.57	10.35	3.50
MicFormer⁃HMD	88.23	10.13	3.34

表2 MicFormer?HMD消融实验结果

Table 2 Ablation study results of MicFormer-HMD

配置	Dice相似系数/%	95%豪斯多夫距离/mm	平均对称表面距离/mm
MicFormer	85.57	10.35	3.50
+DSConv	86.52	10.23	3.34
+MBFA	87.47	10.20	3.25
+HGM	88.23	10.13	3.34

图6 模型在不同高斯噪声水平下的分割性能

Fig.6 Segmentation performance of models under different Gaussian noise levels

图7 遮挡掩模尺寸对Dice相似系数的影响

Fig.7 The effect of occlusion mask size on Dice similarity coefficient

表3 MicFormer?HMD在单模态输入下的分割性能对比

Table 3 Comparative analysis on segmentation performance of MicFormer?HMD in single?modal input

输入配置	Dice相似系数/%
输入配置	MicFormer⁃HMD	TransBTS	U⁃Net
CT	80.20	77.30	72.10
MRI	84.70	78.20	74.50
CT+MRI	88.23	83.92	80.49

参考文献 16

[1]	LITJENS G， KOOI T， BEJNORDI B E， et al. A survey on deep learning in medical image analysis［J］. Medical Image Analysis， 2017， 42： 60⁃88.
[2]	HEIMANN T， MEINZER H P. Statistical shape models for 3D medical image segmentation： A review［J］. Medical Image Analysis， 2009， 13（4）： 543⁃563.
[3]	SHEN D G， WU G R， SUK H I， et al. Deep learning in medical image analysis［J］. Annual Review of Biomedical Engineering， 2017， 19： 221⁃248.
[4]	LI Y H， EL HABIB DAHO M， CONZE P H， et al. A review of deep learning⁃based information fusion techniques for multimodal medical image classification［J］. Computers in Biology and Medicine， 2024， 177： 108635.
[5]	GUO Z， LI X， HUANG H， et al. Deep learning⁃based image segmentation on multimodal medical imaging［J］. IEEE Transactions on Radiation and Plasma Medical Sciences， 2019， 3（2）： 162⁃169.
[6]	TIRUPAL T， MOHAN B C， KUMAR S S. Multimodal medical image fusion techniques：A review［J］. Current Signal Transduction Therapy， 2021， 16（2）： 142⁃163.
[7]	WANG H N， CAO P， YANG J Z， et al. Narrowing the semantic gaps in U⁃net with learnable skip connections： The case of medical image segmentation［J］. Neural Networks， 2024， 178： 106546.
[8]	ZHANG Z L， XU C， LI Z P， et al. Multi⁃scale fusion semantic enhancement network for medical image segmentation［J］. Scientific Reports， 2025， 15（1）： 23018.
[9]	QIU Y H， SHAN D D， WANG Y N， et al. A topology⁃preserving three⁃stage framework for fully⁃connected coronary artery extraction［J］. Medical Image Analysis， 2025， 103： 103578.
[10]	ZHAO Z W， ZHANG Z X， LIU Y H， et al. DeformCL： Learning deformable centerline representation for vessel extraction in 3D medical image［C］//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）. Nashville： IEEE， 2025： 30896⁃30905.
[11]	QI Y L， HE Y T， QI X M， et al. Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation［C］//2023 IEEE/CVF International Conference on Computer Vision （ICCV）. Paris： IEEE， 2023： 6047⁃6056.
[12]	FAN X X， LIU L， ZHANG H R. Multimodal information interaction for medical image segmentation［EB/OL］. （2024⁃04⁃25）［2025⁃10⁃23］. https：//arxiv.org/abs/2404.16371.
[13]	RONNEBERGER O， FISCHER P， BROX T. U⁃Net： Convolutional networks for biomedical image segmentation［C］//Medical Image Computing and Computer⁃Assisted Intervention–MICCAI 2015. Cham： Springer， 2015： 234⁃241.
[14]	CAO H， WANG Y Y， CHEN J， et al. Swin⁃unet： Unet⁃like pure transformer for medical image segmentation［C］//Computer Vision–ECCV 2022 Workshops. Cham： Springer， 2023： 205⁃218.
[15]	OKTAY O， SCHLEMPER J， FOLGOC L L， et al. Attention U⁃net： learning where to look for the pancreas［EB/OL］. （2018⁃04⁃11）［2025⁃10⁃25］. https：//arxiv.org/abs/1804.03999.
[16]	WANG W X， CHEN C， DING M， et al. TransBTS： Multimodal brain tumor segmentation using transformer［C］//Medical Image Computing and Computer Assisted Intervention–MICCAI 2021. Cham： Springer， 2021： 109⁃119.