Detection Of Fetal Congenital Heart Disease Based On Multiscale Feature Of Fetal Echocardiogram

Computer-aided diagnosis(CAD)is playing an increasingly important role in medical diagnosis.The efficient and accurate diagnosis of fetal echocardiography for fetal congenital heart disease is very important for early detection and treatment.The traditional automatic diagnosis of adult heart is often based on the analysis and judgment of certain standard planes of the heart.However,due to the poor quality of fetal echocardiogram,the complicated structure and the challenging work of obtaining the standard planes automatically and accurately,we could not follow the traditional method to get a good diagnosis result.To solve this problem,this paper presents a new and complete automatic diagnostic system for fetal echocardiogram.We used the fetal echocardiogram of the original 4D(3D + T)images and the video sequence of standart planes,respectively,to establish their own automatic diagnose system,and combine the two methods to form a complete system to meet the needs of doctors in the clinic.In this paper,the 3D static features and 3D motion features of the images were extracted with 3D SIFT and 3D HOF respectively from the original 4D image data,with the fuzzy imaging of the fetal heart and the difficulty of the acquisition of standard plane.The One-Class SVM classifier is used to classify the data,and a good classification diagnosis result is achieved.The false negative rate is 0 for the diagnosis of abnormal images.In the view of the fact that the diagnosis of fetal diseases by the method of whole 4D image can not obtain the details of CHD,a new fetal echocardiography automatic diagnose system based on planes is proposed in this paper.After collecting the training data of four standard planes(four-chamber view,abdominal cross-section,three-vessel section,and short-axis section of the aorta)from the acquired sequence images by artificial methods,we extract the feature of the videos and train One-Class SVM classifier to determine whether a plane is standard plane or not.And then we get the proportion of the standard plane in the whole image planes,as the new feature vector to train Adaboost classifier.Finally,we could get the classification of the image.The diagnostic effect is worse than the previous method,but it can help determine the image of abnormal parts,providing more details of the disease information.Finally,this paper used Naive Bayes method combined with the two methods to get the final diagnosis results,improved the credibility of the diagnostic results,and gave a more reasonable way to use the diagnostic system.