
浏览全部资源
扫码关注微信
1.江苏大学,计算机科学与通信工程学院,镇江 212000
2.江苏大学附属医院影像科,镇江 212001
于子腾,Email:2222308005@stmail.ujs.edu.cn,研究方向:人工智能,医学图像处理。
曹雄锋,Email:caoxiongfeng512@163.com。
收稿日期:2025-04-08,
网络出版日期:2025-09-03,
移动端阅览
于子腾,宋雪桦,袁昕等.AMAF-Net:基于自适应多尺度注意力融合网络的冠状动脉精准分割研究[J].重庆医科大学学报,
Yu Ziteng,Song Xuehua,Yuan Xin,et al.Precise coronary artery segmentation based on an adaptive multi-scale attention fusion-network[J].Journal of Chongqing Medical University,
于子腾,宋雪桦,袁昕等.AMAF-Net:基于自适应多尺度注意力融合网络的冠状动脉精准分割研究[J].重庆医科大学学报, DOI:10.13406/j.cnki.cyxb..
Yu Ziteng,Song Xuehua,Yuan Xin,et al.Precise coronary artery segmentation based on an adaptive multi-scale attention fusion-network[J].Journal of Chongqing Medical University, DOI:10.13406/j.cnki.cyxb..
目的
2
基于3D-UNet构建一种自适应多尺度注意力融合网络(adaptive multi-attention fusion-network,AMAF-Net),以提升3D-UNet网络的冠状动脉计算机断层扫描血管成像(computed tomography angiography,CTA)图像分割精准性。
方法
2
在3D-UNet编码器中引入多尺度膨胀卷积模块(multi-scale dilated convolution,MSDC)和压缩与激励(squeeze-and-excitation,SE)注意力机制;在模型架构的瓶颈层中采用标准3D卷积模块;解码器中加入加性注意力机制(additive attention mechanism,AAM),最终构建AMAF-Net。采用公开自动化冠状动脉分割(automated segmentation of coronary artery,ASOCA)数据集和自制冠状动脉CTA40数据集中的冠状动脉CTA图像作为AMAF-Net及8种经典U-Net网络模型的冠状动脉分割训练、验证和测试数据。上述网络模型的冠状动脉分割性能评价包括指标评价(Dice相似系数、召回率、精确率和平均表面对称距离)、视觉评价和消融实验。不同网络模型的分割性能差异比较采用单因素方差分析及Turkey检验。
结果
2
在ASOCA数据集和CTA40数据集中,9种模型间Dice相似系数、召回率、精确率和平均表面对称距离差异皆具有统计学意义(均
P
<
0.001)。AMAF-Net的Dice相似系数、精确率和平均表面对称距离最高,召回率相对排名靠前,且各指标与多个模型间差异具有统计学意义(均
P
<
0.05)。AMAF-Net在冠状动脉复杂结构区域皆具有最佳的分割效果,且所有分
割结果最接近于手动分割金标准真实标签。消融实验中,随着模块数目增加,网络模型的Dice相似系数、召回率、精确率逐渐增加,平均表面对称距离逐渐降低,各模型间所有指标间差异皆具有统计学意义(均
P
<
0.001)。
结论
2
AMAF-Net的冠状动脉分割精准性得到明显提升,且表现出良好泛化性,具有潜在临床推广应用价值。
Objective
2
To construct an adaptive multi-scale attention fusion-network(AMAF-Net) based on 3D-UNet,and to improve the precision of the segmentation of coronary computed tomography angiography(CTA) images using 3D-UNet network.
Methods
2
The multi-scale expanded convolution module(MSDC) and the Squeeze-and-Excitation(SE) attention mechanism were introduced into the 3D-UNet encoder. Standard 3D convolutional modules are adopted in the bottleneck layer of the model architecture. The additive attention mechanism(AAM) is added to the decoder,and finally the AMAF-Net is constructed. The coronary CTA images in the public dataset automated segmentation of coronary artery(ASOCA) and/or the self-made dataset CTA40 were used as the training,validation,and test data for coronary artery segmentation using AMAF-Net and eight classical U-Net network models. The performance of the above network models in coronary artery segmentation was assessed based on Dice similarity coefficient,recall,precision,average surface symmetry distance,visual evaluation,and ablation experiment. A one-way analysis of variance and the Turkey test were used to compare the segmentation performance of different network models.
Results
2
In the ASOCA and CTA40 datasets,there were significant differences in Dice similarity coefficient,recall,precision,and average surface symmetry distance between the nine models(all
P
<
0.001). AMAF-Net had the highest Dice similarity coefficient,precision,and average surface symmetry distance and a relatively high recall,and there were significant differences in each indicator between AMAF-Net and various other models(all
P
<
0.0
5). AMAF-Net had the best segmentation performance in the negative complex structure region of the coronary artery,and all segmentation results were the closest to the gold standard of manual segmentation true label. In the ablation experiment,with the increase in the number of modules,the network model showed gradual increases in Dice similarity coefficient,recall,and precision and a gradual reduction in average surface symmetry distance,and there were significant differences in all indicators among the models(all
P
<
0.001).
Conclusion
2
There is a significant improvement in the accuracy of coronary artery segmentation based on AMAF-Net,with good generalization and a potential clinical application value.
Maria T , Elias D . Long-Term Coronary Artery Disease Risk Prediction with Machine Learning Models [J ] . Sensors , 2023 , 23 ( 3 ): 1193 - 1193 .
Brown JC , Gerhardt TE , Kwon E . Risk Factors for Coronary Artery Disease . In: StatPearls , 2023 , StatPearls Publishing : 1 - 129 .
Ronneberger O , Fischer P , Brox T . U-Net:Convolutional Networks for Biomedical Image Segmentation [C ] Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015 , 2015 ,Cham. Springer International Publishing : 234 - 241 .
Çiçek Ö , Abdulkadir A , Lienkamp SS , et al . 3D U-Net:Learning Dense Volumetric Segmentation from Sparse Annotation [J ] . CoRR , 2016 ,abs/1606.06650.
Wang W , Chen C , Ding M , et al . TransBTS:Multimodal Brain Tumor Segmentation Using Transformer [C ] Medical Image Computing and Computer Assisted Intervention-MICCAI 2021 , 2021 ,Cham. Springer International Publishing : 109 - 119 .
Milletari F , Navab N , Ahmadi SA . V-Net:Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation [J ] . CoRR , 2016 ,abs/1606.04797.
Badrinarayanan V , Kendall A , Cipolla R . SegNet:A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation [J ] . IEEE Transactions on Pattern Analysis and Machine Intelligence , 2017 , 39 ( 12 ): 2481 - 2495 .
Gu R , Wang GT , Song T , et al . CA-Net:Comprehensive Attention Convolutional Neural Networks for Explainable Medical Image Segmentation [J ] . IEEE Transactions on Medical Imaging , 2020 , 40 ( 2 ): 699 - 711 .
Mou L , Zhao YT , Fu HZ , et al . CS 2 -Net:Deep Learning Segmentation of Curvilinear Structures in Medical Imaging [J ] . Medical Image Analysis , 2021 , 67 : 101874 .
Song AL , Xu LS , Wang L , et al . Automatic Coronary Artery Segmentation of CCTA Images with an Efficient Feature-Fusion-and-Rectification 3D-UNet [J ] . IEEE Journal of Biomedical and Health Informatics , 2022 , 26 ( 8 ): 4044 - 4055 .
Yu L , Cheng JZ , Dou Q , et al . Automatic 3D Cardiovascular MR Segmentation with Densely-Connected Volumetric ConvNets [C ] Medical Image Computing and Computer-Assisted Intervention-MICCAI 2017 , 2017 ,Cham. Springer International Publishing : 287 - 295 .
Dong C , Dai D , Li Z , et al . A Novel Deep Network with Triangular-Star Spatial-Spectral Fusion Encoding and Entropy-Aware Double Decoding for Coronary Artery Segmentation [J ] . Information Fusion , 2024 , 112102561 - 102561 .
0
浏览量
0
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621