Right Ventricular Magnetic Resonance Image Segmentation Based On Anatomical Feature Landmarks

Nowadays,medical image segmentation has become an important image segmentation technology.Quantitative evaluation of the function of the medical image of the right ventricle has very important guiding significance for the diagnosis,treatment and prognosis of cardiac diseases such as infarct heart failure and myocardial hypertrophy.This article combines the knowledge of the anatomy of the cardiac,uses the level set theory to achieve accurate segmentation of the right ventricle,proposes an improved model based on the U-Net network,builds a cardiac key landmark detection network,and finally uses the public data set to analyze the segmentation method and proposed the key landmark detection network is verified by experiments,and several key evaluation indicators show that the algorithm in this paper is effective.The initialization of the level set function is an important task.Choosing a good initialization can reduce the iteration time of the level set and increase the evolution speed of the level set.According to several mainstream methods of image segmentation,a preliminary segmentation plan is proposed,and a simple method is used to construct the initial level set.The trabecula and papillary muscles are attached to the endocardium.According to the medical imaging results,the gray scale of the trabecula and other structures is similar to that of the myocardium.During the segmentation process,the contours of these structures will be mistakenly regarded as the endocardium,leading to segmentation.The accuracy of the result is low.This article improves the convex-preserving level set method and segmenting the right ventricle.Because the right ventricle is partially concave inward,part of the curve of the right ventricle will leak during the segmentation process using the convex-preserving model.In response to this problem,a position constraint is introduced into the level set energy term to constrain the evolution range of the right ventricle,and the algorithm verified by the public data set can accurately segment the right ventricle.The acquisition of level set initialization has always been a difficult problem,and manual intervention is inevitable in the segmentation process.Therefore,the level set method is difficult to achieve automatic segmentation.In response to this problem,this paper makes improvements based on the U-Net model,and proposes an accurate key landmark detection technology,which uses a pixel-level classifier to detect key landmark regions,and then uses the detected key landmark position information combined with morphological methods obtain the key area,automatically intercept the cardiac area for initial segmentation.The initial segmentation result is multiple connected areas including the right ventricular area.The key landmark information is again used to obtain the right ventricular area to construct the initial level set function,and finally the right ventricular area is accurately segmented.The introduction of key landmarks realizes the automatic segmentation of the right ventricle.