摘要
精确分析患者的心功能状态,并尽早地、准确地诊断出心脏疾病,对提高该病的治疗效果,降低医疗成本有着重大意义.众多影像手段中,心脏磁共振的软组织对比度最高,但是心脏核磁数据序列多、融合难,处理需要专业医师人工勾画,非常耗时.因此,本文基于多任务学习机制,构建双支深度神经网络,结合卷积神经网络及循环神经网络,提取心脏核磁的空间及运动特征,在训练样本量有限的情况下实现了对正常心脏、扩心病、肥心病的准确诊断.本文方法与经典算法C3D及LRCN对比获得了更高的识别准确率,综合AUC值均达到了0.94以上.
Accurate analysis of the patient's heart function and early diagnosis of a myocardial disease can greatly improve the effect of treatment and reduce the medical cost significantly.Among medical imaging technologies,cardiac magnetic resonance(CMR)has the highest tissue contrast.However,CMR has numerous sequences and makes data fusion difficult.It is very time-consuming to pre-process CMR data manually.Therefore,based on the multi-task learning mechanism,this paper constructs a dual-branch deep neural network,combines convolutional neural network and recurrent neural network.Then the spatial and motion characteristics of cardiac MRI are extracted to recognize normal group,dilated cardiomyopathy,and hypertrophic cardiomyopathy with limited data.In the experiment,compared to C3D and LRCN,the proposed method obtains the best performance.Both accuracy and AUC score achieved 0.94.
作者
肖晶晶
崔春
陈佳
XIAO Jingjing;CUI Chun;CHEN Jia(Department of Medical Engineering, The Second Affiliated Hospital of the Army Medical University, Chongqing 400037, China;Department of Radiology, The Second Affiliated Hospital of the Army Medical University, Chongqing 400037, China)
出处
《测试技术学报》
2020年第5期390-395,共6页
Journal of Test and Measurement Technology
基金
国家自然科学基金资助项目(61701506)
临床科研人才计划资助项目(2018XLC3023)。
关键词
心脏核磁共振
心肌病识别
多任务学习
双支网络
卷积网络
cardiac magnetic imaging
cardiomyopathy recognition
multitask learning
dual-network
convolutional network
