Automated Detection And Segmentation Of The Left Ventricle In Cardiac Mri Images Based On Distance Metric And Deep Learning

The left ventricle of the heart,as the core blood supply tissue of human organs,can provide important basis for the diagnosis of heart diseases by measuring many parameters such as ventricular volume,ejection fraction and myocardial thickness.In order to accurately measure the key parameters of the left ventricle,traditional methods rely on the manual marking of the left ventricle by experienced anatomical experts to achieve parameter calculation.However,manual marking is tedious and troublesome.Therefore,it is necessary to develop an automatic left ventricular detection,localization and segmentation model for auxiliary evaluation the key parameters of the left ventricle.The paper is based on distance metric learning and deep learning to detect,locate and segment left ventricular tissiue for cardiac MRI images.In terms of left ventricular detection,a left ventricular detection method is proposed by combining candidate region two-level distance metric learning and CNN classification and regression joint learning.In the candidate region generation stage,the super-pixel method is employed to generate the initial region and further merged into the intermediate region.The supervised two-level distance metric learning algorithm is designed to fuse the intermediate regions to construct the target candidate regions.In the detection stage,the approach of joint learning with CNN classification and regression is employed to locate candidate regions.In terms of left ventricular landmark localization,the paper proposed a left ventricular landmark localization and identification method in cardiac MRI based on distance metric learning and deep CNN regression that includes two stages.In the sample generation stage,a dual-channel salient patch mining module is designed by combining super-pixel and grid image where patch feature representation is extracted by the embedded triplet network.In the regression localization stage,the weight of the embedded triplet network is shared to build the CNN regression model,and the salient patches are employed to predict the landmark coordinate point cloud clusters.In terms of left ventricular segmentation,the paper proposed a left ventricular multi-atlas segmentation method combining distance metric learning and deep SSAE features,which is composed of initial rough segmentation stage and boundary fine segmentation stage.In the initial rough segmentation stage,the label fusion module of majority voting was used to predict the rough segmentation map of myocardial tissue.In the boundary fine segmentation stage,the SSAE network is used to learn the deep patch feature representation of the myocardial boundary region,the distance metric learning and the weighted label fusion module is used to evaluate the low-confidence boundary region of the rough segmentation map and output the final myocardial segmentation results.The proposed method is performed on the Cardiac Atlas Project(CAP)dataset and results validate the reasonability of module settings in proposed method.Further experiments show that the proposed method achieves competitive detection,localization and segmentation accuracy.