Study On Facial Assistant Diagnosis Method For Adenoid Hypertrophy

Adenoids are called adenoid hypertrophy if they continue to swell for some reason,affecting nasal breathing and physical health.Adenoid hypertrophy is a common childhood disease.At present,the diagnosis method of adenoid hypertrophy is mainly nasal endoscopy,and the invasive diagnosis of nasal endoscopy will make patients feel uncomfortable and painful.It is inconvenient and has a low penetration rate,especially not suitable for remote areas where medical resources are scarce.This paper proposes a facial auxiliary diagnosis research method for adenoid hypertrophy,which can be non-invasive,painless,suitable for remote areas and places with poor medical resources,expand the screening range of adenoid hypertrophy,and facilitate the diagnosis process.The main innovation points and characteristics of this paper are:(1)Innovation in research field: a kind of adenoid hypertrophy facial auxiliary diagnosis auxiliary diagnosis method is proposed.(2)extract local mixed features according to the standards of medical experts: in the process of extracting local features,image segmentation based on coordinate points is carried out for the lip region.(3)feature extraction method innovation: the feature selection of global geometric features and global texture features is carried out by using XGBoost algorithm,and a feature set with high importance is obtained for feature mixing.Compared with the global feature extraction method without feature selection,this method has shorter time and better performance than the trained classifier.In this paper,the data from the Beijing children's hospital sleep apnea center,by comparing the feature extraction method training generated classifier can be seen: globaltexture feature and texture characteristics of the hybrid training generated classifier,accurate rate was 96%,the recall rate is 96%,F1 value 96%,therefore,the global geometric features and global texture feature extraction method is superior to extract only the global features of hybrid and only local mixed feature extracting method,the method to get the classifier performance is better.