Intelligent Analysis Of Myocardial Infarction Image Based On Deep Learning

Delayed enhanced cardiac magnetic resonance(DE-CMR)is considered the gold standard for assessing myocardial infarction and activity and is commonly used to diagnose cardiac disease.Identification of myocardial infarction region from image and quantitative analysis of scar can provide important parameter information,so accurate segmentation of infarction region is crucial for clinical diagnosis and prognosis of patients.At present,in clinical practice,doctors first screen images containing myocardial infarction,and then manually segment the scar contour to quantitatively evaluate the degree of myocardial infarction.The work is often time-consuming and cumbersome,and standards vary from doctor to doctor,which can lead to inaccurate segmentation results.To meet the above requirements,this paper designed the myocardial lesion classification network and the myocardial infarction automatic segmentation network based on deep learning method,to realize the intelligent analysis of cardiomyopathy imaging.In this paper,after studying various existing medical image classification and segmentation methods at home and abroad,a cardiomyopathy image classification network is proposed.The network is implemented based on migrating Resnet50.The network learns the superficial features of images through pre-training,and then learns the high-level semantic features from DE-CMR images to update the output layer of the network.An image enhancement method was designed based on GAN network,which was used to synthesize myocardial MR images.The training data set of the classification network was expanded,and the classification accuracy of the network was improved.The classification accuracy of the test set reached 93.83%.In addition,this paper proposes an automatic segmentation method for cardiac scar based on deep learning method.The method adopts a cascade scheme,which firstly detects the boundary box and extracts the regions of interest around the myocardium,and then segments the myocardium and cardiac scar successively through two segmentation networks.The segmentation network is designed based on the fully convolutional DenseNet network architecture.The residual path is added in the hop connection to reduce the semantic gap between the encoder and decoder.The multi-scale residual module is introduced in the input layer to expand the model’s receptive field and enhance the multi-scale feature extraction capability of the network.The model in this paper was trained and predicted based on the data of EMIDEC and Myops2020 challenge.Experimental results show that the method of accurate classification of cardiomyopathy images and automatic segmentation of cardiac scar are realized in this paper.Comparing with other algorithms and experimental analysis,the effectiveness of the two methods proposed in this paper is verified.