Design And Implementation Of Left Ventricular Volume Analysis System Based On Deep Learning

Cardiovascular diseases have become the leading cause of death worldwide.In the field of clinical diagnosis of cardiovascular diseases,the processing and analysis of medical images play an important role in the diagnosis of diseases.This thesis uses deep learning techniques to complete the segmentation and the volume prediction of the left ventricle based on cardiac magnetic resonance(MRI)images.It may provide a more accurate reference for the diagnosis of heart-related diseases.The main content of the thesis is to use the deep learning technology to predict the volume of the left ventricular cavity,and design and implement a Web system.For the prediction of left ventricular volume,this thesis proposes two methods.One method is based on an U-net network,adding a LSTM(long short-term memory network)to connect the spatial information of the short-axis MRI slices of the ventricle to achieve left ventricular segmentation,and then the volume is calculated.This method can meet the basic requirements of left ventricular segmentation and left ventricular volume prediction.Another method is to use a multi-scale atlas matching method to achieve the accurate localization of the left ventricular region.And then a three-dimensional convolution neural network was used to predict the left ventricle volume.Experimental results show that the volume prediction error of this method is significantly lower than that of the two-dimensional convolutional neural networks.Finally,the Flask technology was used to develop the left ventricular cavity volume analysis system.The system uses the proposed algorithm to provide users with functions such as left ventricular segmentation and left ventricular volume prediction.It supports users to upload data for related analysis and fill in corresponding diagnostic opinions.The system has good performace such as higher accuracy and better ease of use.