基于卷积神经网络的隐匿性旁路预测模型

Model for Predicting Concealed Accessory Pathway Based on Convolutional Neural Network

隐匿性旁路(CAP)是一种引起心跳突然加速、心悸和胸闷的心脏疾病。针对目前临床医师尚无法通过窦性心律心电图(ECG)对隐匿性旁路进行诊断的现状,基于临床病例建立包含隐匿性旁路患者术前窦性心律心电图及健康对照人群心电图的数据集,并提出一种以ResNet26为基线网络的利用窦性心律心电图自动识别预测隐匿性旁路患者的卷积神经网络CAPNet。创建初始模块(IB),提升模型非线性表达能力。引入非对称卷积以改进瓶颈残差模块,更好地捕捉心电特征的水平和垂直方向信息,丰富特征空间。使用注意力机制,加强模型对心电图中重点波段区域的关注。实验结果表明,CAPNet模型的预测性能优于对比的经典卷积神经网络模型,与ResNet26相比,F1值、准确率、灵敏度和精确率分别提升了2.41、0.89、4.34和0.47个百分点。上述实验结果验证了CAPNet模型的有效性与优越性。

A Concealed Accessory Pathway(CAP)is a heart condition characterized by rapid heartbeat,palpitations,and shortness of breath.However,clinicians cannot currently diagnose CAPs via sinus rhythm Electrocardiogram(ECG).Based upon clinical cases,the present study establishes a dataset containing the preoperative sinus rhythm ECG data of healthy subjects and patients with the CAP,and proposes a Convolutional Neural Network(CNN)based on ResNet26,referred to as CAPNet,to automatically identify and predict the CAP using via the sinus rhythm ECG.An Initialization Block(IB)is established to improve the nonlinear expression of the model.An asymmetric convolution block is introduced into the bottleneck residual block to better capture the horizontal and vertical directional information of the ECG features,allowing the module to enrich the feature space.Furthermore,an attention mechanism is used to enhance the attention of the model to the key band region in the ECG.The results demonstrate that CAPNet model outperforms CNN models in predicting the CAP.The common indicators of CAPNet model including the F1 score,accuracy,sensitivity,and precision increase by 2.41,0.89,4.34,and 0.47 percentage points,respectively.These experimental results validate the effectiveness and superiority of the CAPNet model.