摘要
心脏磁共振检查是用于评估心脏结构和功能的一种非侵入式的医学成像技术,与其他医学成像技术相比,不存在辐射伤害并且更擅长捕捉软组织细节,可为医生提供关于心脏结构和功能的详细信息,在心脏疾病的诊断和治疗中发挥着至关重要的作用。为了精准分割心脏磁共振图像(Magnetic resonance image,MRI),在nnU-Net自适应分割框架的基础上提出基于改进nnU-Net的分割方法。通过在编码器部分应用残差模块代替原始卷积以缓解梯度消失问题并增强特征学习,利用在最底层瓶颈部分引入十字交叉注意力模块以捕获长距离依赖关系并提升模型的特征表达能力,此外,在跳跃连接部分加入卷积块注意力模块以减小噪声干扰并聚焦于关键特征。在心脏自动诊断挑战(Automatic cardiac diagnosis challenge,ACDC)数据集上进行实验,结果表明基于改进nnU-Net的分割方法具有更精确的分割效果。
Cardiac magnetic resonance examination is a non-invasive medical imaging technology used to evaluate the structure and function of the heart.Compared with other medical imaging technologies,it does not cause radiation damage and is better at capturing soft tissue details,providing doctors with detailed information about the structure and function of the heart.Detailed information plays a vital role in the diagnosis and treatment of heart disease.To accurately segment cardiac Magnetic resonance image(MRI),a segmentation method based on improved nnU-Net is proposed based on the nnU-Net adaptive segmentation framework.By applying a residual module instead of the original convolution in the encoder section to mitigate the gradient vanishing problem and enhance feature learning,a criss-cross attention module is introduced in the bottleneck part of the bottom layer to capture the long-distance dependencies and enhance the feature representation ability of the model,and in addition,a convolutional block attention module is added to the hopping connection part to reduce the noise interference and focus on the key features.Experiments are conducted on the Automatic cardiac diagnosis challenge(ACDC)dataset,and the results show that the segmentation method based on the improved nnU-Net has more accurate segmentation results.
作者
刘佳悦
孔凡辉
马吉权
LIU Jiayue;KONG Fanhui;MA Jiquan(School of Computer Science and Technology,Heilongjiang University,Harbin 150080,China;School of Data Science and Technology,Heilongjiang University,Harbin 150080,China)
出处
《黑龙江大学自然科学学报》
CAS
2024年第5期597-605,共9页
Journal of Natural Science of Heilongjiang University
基金
黑龙江省自然科学基金资助项目(LH2021F046)
工业控制技术全国重点实验室开放课题(ICT2023B13)。
