[1]王建军,李婉晴*,张 敏.多尺度空洞 U-Net 网络的电影 CMR 图像分割[J].计算机技术与发展,2022,32(11):177-182.[doi:10. 3969 / j. issn. 1673-629X. 2022. 11. 026]
　WANG Jian-jun,LI Wan-qing*,ZHANG Min.Multi-scale Dilated U-Net Network for Cine CMR Image[J].,2022,32(11):177-182.[doi:10. 3969 / j. issn. 1673-629X. 2022. 11. 026]
点击复制

多尺度空洞 U-Net 网络的电影 CMR 图像分割()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年11期
页码:
177-182
栏目:
人工智能
出版日期:
2022-11-10

文章信息/Info

Title:
Multi-scale Dilated U-Net Network for Cine CMR Image
文章编号:
1673-629X(2022)11-0177-06
作者:
王建军; 李婉晴* ; 张 敏
河北经贸大学 信息技术学院,河北 石家庄 050061
Author(s):
WANG Jian-jun; LI Wan-qing* ; ZHANG Min
School of Information Technology,Hebei University of Economics and Business,Shijiazhuang 050061,China
关键词:
心脏核磁共振图像; U-Net 网络; 迁移学习; 多尺度空洞卷积; 特征融合
Keywords:
cardiac magnetic resonance image; U-Net network; transfer learning; multi-scale dilated convolution; feature fusion
分类号:
TP301
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 11. 026
摘要:
在电影心脏核磁共振(CMR)图像上准确分割左心室、右心室和心肌是心脏功能评估和诊断的重要步骤。 然而,大多数带标注的 CMR 图像数据量较少,无法满足训练需求,同时 CMR 图像中心脏结构复杂,心室及心肌边界不清晰,导致分割效果欠佳。 因此,该文提出了一种基于迁移学习和多尺度空洞 U-Net 网络的 CMR 图像分割方法,使用迁移学习,将预训练模型得到的网络参数迁移到目标模型上作为目标模型的初始化参数,提高网络的特征学习能力,解决 CMR 图像数据量不足的问题;在 U-Net 网络中引入多尺度空洞卷积模块,使用空洞卷积代替普通卷积在参数不变的情况下扩大了感受野,并且采用多尺度特征融合提取更加精细的特征,解决 CMR 图像边界曲线欠分割的问题。 实验结果表明,该方法能有效实现心脏中左心室、右心室和心肌的准确分割,平均 Dice 系数和 Hausdorff 距离平均值分别为 0. 902 和 4. 219 mm,对比其他网络分割模型明显提高了分割精度。
Abstract:
Accurate segmentation of the left ventricle,right ventricle and myocardium on cine cardiac magnetic resonance ( CMR) imagesis an important step in cardiac function assessment and diagnosis. However,most CMR images with annotation have less data to meet thetraining requirements,while the complex heart structures and unclear boundaries of ventricles and myocardium in CMR images lead topoor segmentation results. Therefore,we propose a CMR image segmentation method based on transfer learning and multi-scale dilatedU-Net network. According to transfer learning,the network parameters obtained from the pre-trained model are migrated to the targetmodel as the initialization parameters of the target model to improve the feature learning ability of the network and solve the problem ofinsufficient data volume of CMR images. The multi-scale dilated convolution module is introduced in the U-Net network. The use ofdilated convolution instead of ordinary convolution expands the receptive field with the same parameters,and multi-scale feature fusion isused to extract finer features to solve the problem of under-segmentation of CMR image boundary curves. The experiment shows that theproposed method can effectively achieve accurate segmentation of the left ventricle,right ventricle and myocardium in the heart,and theaverage Dice correlation coefficient and Hausdorff distance averages are 0. 902 and 4. 219 mm,respectively,which significantly improvethe segmentation accuracy compared with other network segmentation models.

相似文献/References:

[1]刘 军,孙 庆,刘 玮,等.基于 U-Net 网络和椭圆度量学习的防震锤锈蚀识别[J].计算机技术与发展,2020,30(11):163.[doi:10. 3969 / j. issn. 1673-629X. 2020. 11. 030]
　LIU Jun,SUN Qing,LIU Wei,et al.Identification of Anti-vibration Hammer Corrosion of High-voltage Transmission Lines Based on U-Net Network and Elliptic Metric Learning[J].,2020,30(11):163.[doi:10. 3969 / j. issn. 1673-629X. 2020. 11. 030]
[2]李松宇.基于 HED—UNet 遥感图像建筑物语义分割[J].计算机技术与发展,2022,32(S2):58.[doi:10. 3969 / j. issn. 1673-629X. 2022. S2. 010]
　LI Song-yu.Semantic Segmentation of Buildings in Remote Sensing Images Based on HED-UNet[J].,2022,32(11):58.[doi:10. 3969 / j. issn. 1673-629X. 2022. S2. 010]
[3]金海燕,张 锦,王海鹏,等.基于控制点特征学习的前列腺组织轮廓线提取方法[J].计算机技术与发展,2023,33(07):27.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 004]
　JIN Hai-yan,ZHANG Jin,WANG Hai-peng,et al.Extraction of Prostate Tissue Contour Based on Control Points Feature Learning[J].,2023,33(11):27.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 004]

更新日期/Last Update: 2022-11-10