Research On 3D Face Auxiliary Diagnosis Method For Acromegaly

Acromegaly is a group of disorders with excessive growth hormone,mainly caused by pituitary adenomas that secrete GH,resulting in excessive tissue growth that persists for many years.Clinical manifestations of acromegaly Typical facial tissue features are: deep nasolabial groove,enlarged nose or brow,protrusion of jaw,and protrusion of zygomatic arch.Because facial changes occur so mysteriously,they are often overlooked,especially by the patient or by people who frequently see the patient,to the detriment of disease control.In this study,the research process of assisted diagnosis and automatic classification was carried out based on the 3D model of patients.This paper strictly followed the clinical standards of acromegaly,and from the perspective of clinical significance,the key technologies for automatic classification of 3D model of patients were studied.Main research contents and innovation points of this paper:(1)Combined with the existing 3D point cloud processing technology,a processing method more suitable for the face is selected to preprocess the 3D point cloud,including point cloud smoothing processing,face cutting and face correction.(2)Combined with the nose tip location method of 3D face and according to the facial features of acromegaly,a more targeted face key point extraction method was selected;Through key points,global geometric features and local regional curvature features of patients’ faces were extracted.(3)The accuracy of 2D and 3D geometric features in disease recognition was compared.The recognition effects of SVM,Adaboost and XGBoost on curvature features were compared.The experimental data used in this paper were collected from the Department of Neurology of Peking Union Medical College Hospital between July 2018 and September2020.Among them,157 facial 3D models of acromegaly were made.There were 263 age-and sex-matched patients in the control group.By preprocessing the data,the global and local features are extracted and classified by the classification algorithm.The experimental results show that the accuracy of XGBoost classifier based on geometric features is slightly higher than that of SVM classification algorithm.By comparing the recognition accuracy of 2D and3 D geometric features,it is proved that 3D geometric features have advantages in the recognition of acromegaly,while the overall recognition effect of curvature features is weaker than that of geometric features.