Research On Automatic Segmentation And Auxiliary Diagnosis Of Cardiac Magnetic Resonance Image Based On Deep Learning

Cardiac magnetic resonance imaging(MRI)is considered the "gold standard" to judge the structure and function of the heart.Accurate acquisition of intracardiac structure is the first step to calculate cardiac clinical indicators,which provides guidance for the diagnosis of cardiovascular diseases.In clinical practice,doctors need to manually draw the contours of left ventricle,right ventricle and myocardium.However,manual segmentation is time-consuming and laborious,and there are differences in standards,which can lead to inaccurate segmentation results vary from person to person.Therefore,a fast,accurate,repeatable and fully automated heart segmentation algorithm is needed.Clinical parameters can be calculated to assist in the diagnosis of cardiovascular diseases after accurate segmentation.In this paper,after research on the existing segmentation algorithms of cardiac MRI image at home and abroad,a short axis cardiac MRI automatic segmentation algorithm based on deep learning is proposed,and a cardiac MRI automatic segmentation and auxiliary diagnosis model is designed.The short axis cardiac MRI automatic segmentation algorithm based on deep learning proposed in this paper is improved based on the network structure of DenseNet and U-net.The automatic segmentation and auxiliary diagnosis model of cardiac MRI firstly segments the input cardiac MRI automatically.Automatic segmentation method includes two steps:preprocessing and segmentation.Preprocessing includes:firstly,region of interest(ROI)detection is performed on short axis cardiac MRI data,Canny edge detection and Hough transform are used,then data standardization and data enhancement are performed.The improved semantic segmentation model based on DenseNet and U-Net is used to segment the ROI results,and the combination of weighted cross entropy loss,dice loss and L2 regular-ization is used in the loss function.According to the segmentation results,the relevant clinical parameters are calculated,the feature extraction is carried out,and the classifier is designed for pathological classification.The experiment of this paper is based on the data set of the Automated Cardiac Diagnostic Challenge(ACDC)held in 2017.The experimental results show that this paper realizes the accurate automatic segmentation and auxiliary diagnosis method of heart region of interest.The proposed method successfully segmented left ventricle,right ventricle and myocardium from cardiac MRI.Compared with other leading segmentation methods in the challenge,excellent results were obtained,which verified the effectiveness of the proposed method.In addition,the designed classifier achieves 94% accuracy on the test data set.