Research On Segmentation And Classification Of Multi-source Cardiac Medical Images Based On Multi-scale Convolution

The booming development of computer-aided diagnosis and medical imaging technology has facilitated rapid diagnosis and effective treatment of various diseases.The high incidence of cardiovascular disease in China is easy to cause sudden cardiac death,which seriously threatens life and health.In recent years,the convolutional neural network approaches have made remarkable achievements in the field of computer vision.Convolutional neural network can improve the efficiency of diagnosis and the quality of life of patients.This paper begins by describing the research significance of the topic from the current social and technological context.The research and application of convolutional neural network in medical imaging,especially in cardiac magnetic resonance imaging are fully discussed.An introduction of the basic technical principles and details,it provides the theoretical basis and practical guidance for the work of this paper.In this paper,the work of computer-aided diagnosis using cardiac magnetic resonance imaging is as follows:(1)For the prognostic task of cardiac patients,this paper designs a convolutional neural network which receives two modalities of cardiac magnetic resonance image data and clinical record data.And to effectively predict the prognostic risk of multi-source cardiology images by multimodal fusion techniques and loss function optimization for unbalanced samples.(2)For the segmentation task of cardiac anatomical tissue of clinical interest.,this paper designs a multi-scale convolution module to recover fine spatial detail information using depth-to-space based up-sampling algorithms for segmentation of anatomical structures such as myocardium and blood pool.(3)Using the convolutional neural network model with the encoder-decoder structure,the classification of the dilated cardiomyopathy and segmentation of the myocardium was completed using an encoder output classification label while the decoder output segmentation mask.