The Research Of Automatic Facial Classification For Turner Syndrome

Turner syndrome(TS),one of the most common chromosomal disorders,results from the total or partial absence or structural abnormality of one copy of X chromosome.The prevalence of TS in live-born girls is approximately 1/2500.The diagnostic method of Turner syndrome is to see if there is a typical clinical feature(such as short stature)by an experienced doctor in the early stage,and then confirmed by chromosome karyotype test.Traditional diagnostic methods have many shortcomings,such as long diagnosis period and high cost,which lead to delay in treatment time.This paper mainly through the analysis of Turner syndrome patients with facial characteristic clinical features,combined with facial recognition technology related content,proposed a set of research process of the use of Turner syndrome patients with 2D facial image auxiliary diagnosis and automatic classification.This research has important research value and clinical significance for the application of face images in the field of medical assistant diagnosis.In this paper,the key techniques for automatic classification of human face images with Turner syndrome were studied in accordance with the clinical standards and clinical significance.The main research content and innovation point of this paper are as follows:(1)Combined with face recognition and digital image processing correlation technology,based on the existing research,we choose a more suitable face image processing method.(2)According to the characteristic facial features of the patient,the suitable face detection,face correction method and feature analysis are used to establish the 68-feature-point model,and the morphological method is used to enhance the feature of the face image.(3)In combination with the 68 feature point model,the global geometric feature,global texture feature and 5 different local characteristics of the patient’s face were extracted,and the features were reduced and integrated.(4)Different classifiers are designed according to different features for automatic classification and detection.The SVM classifier based on geometric features and texture features is designed according to the characteristics of the global feature,and a two-stage Adaboost classifier based on the feature fusion of the matching layer is designed according to the characteristics of the local feature.(5)On the basis of the existing facial dataset,the specific value of the extent of the patient’s facial distortion is preliminarily obtained by causality analysis.This value is of great reference significance for the clinician to make a preliminary diagnosis.The face image used in this paper is collected by Peking Union Medical College Hospital from July 2016 to September 2017.There are 69 patients with Turner syndrome,and 225 cases with age matched contrast images.Through preprocessing the original face image,training the 68-feature-point model and extracting the global features and local features.Finally,automatic classification algorithm is used to identify and classify face images.The best performance of the experiment result up to 84.6%.The experimental results show that it is feasible to analyze and classify facial images of patients with Turner syndrome.