基于CA-MobileNetV2的心肌梗死定位算法研究  

Localization Algorithm of Myocardial Infarction Based on CA-MobileNetV2

作　　者：张鹏飞[1] 叶哲江[1] ZHANG Pengfei;YE Zhejiang(School of Information Engineering and Automation,Kunming University of Science and Technolog,Kunming Yunnan 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500

出　　处：《传感技术学报》2024年第7期1179-1185,共7页Chinese Journal of Sensors and Actuators

摘　　要：为实现临床医疗设备快速辅助诊断心肌梗死(MI)发生的部位。在轻量化卷积神经网络MobileNetV2的基础上结合协调注意力(CA)机制设计出了一种高准确率的MI部位定位算法。从PTB数据集中筛选正常和MI病例的12导联心电图(ECG)样本,将ECG信号进行去噪处理。使用差分阈值法检测出ECG信号的R峰,根据R峰分割出心拍样本,使用心拍数据对所设计模型进行训练和测试。使用准确率、精度、灵敏度、特异性和混淆矩阵对模型的分类性能进行了评估。将训练集迭代60轮后,测试集的准确率达到了99.91%。结果表明,融合CA模块的MobileNetV2模型对于MI部位的定位具有很好的效果,有助于医疗设备实现MI的快速辅助诊断。To achieve a rapid assisted diagnosis of the site of myocardial infarction(MI)occurrence by clinical medical devices,a highaccuracy MI site localization algorithm is designed based on the lightweight convolutional neural network of MobileNetV2 combined with the coordinated attention(CA)mechanism.The 12-lead electrocardiogram(ECG)samples of normal and MI cases are filtered from the PTB dataset,and the ECG signals are denoised.The R-peaks of ECG signals are detected by using the differential thresholding method,the heartbeat samples are segmented according to the R-peaks,and the heartbeat data are used to train and test the model designed.The classification performance of the model is evaluated by using accuracy,precision,sensitivity,specificity and confusion matrix.After iterating the training set for 60 rounds,the accuracy of the test set reached 99.91%.The results show that the MobileNetV2 model incorporating the CA module is effective for localization of MI sites and helps medical devices to achieve rapid assisted diagnosis of MI.

关 键 词：轻量化卷积神经网络 心肌梗死定位 MobileNetV2 注意力机制 心电图 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP183[自动化与计算机技术—计算机科学与技术] R318.6[医药卫生—生物医学工程]