Research On Detection Of Adenoid Hypertrophy In Children Based On Respiratory Airflow

Adenoid hypertrophy(AH)is a common cause of airway obstruction in children the prevalence of AH in children and adolescents is 34.46%.AH is clinically expressed by nasal obstruction,mouth breathing,changes in facial growth,speech problems and behavioral development,nightly snoring,obstructive sleep apnea syndrome,and/or more serious problems.With timely diagnosis and treatment,most sequelae can be avoided or reversed.The more common diagnostic tools used to assess AH are Nasal endoscopy,lateral neck X-ray,and computed tomography(CT),Due to the invasiveness and radiation,it is difficult to assess the changes of adenoids in children frequently by common diagnostic tools.Because of the above clinical pain points,there is an urgent need for a simple,non-invasive and non-radiation new method for the detection and diagnosis of adenoid hypertrophy.For the above clinical challenges,it is urgent for a simple,non-invasive,secure,new method to assess and diagnose AH.This thesis is based on respiratory airflow signals of children with adenoid hypertrophy and healthy children collected at Shandong Provincial Hospital.The purpose of this thesis is to study a simple and non-invasive new method by analyzing the differences in respiratory airflow features between children with adenoid hypertrophy and healthy children.The main research content of this thesis is as follows:(1)Constructing the database of respiratory airflow signals.In this study,rhinomanometer is used as the device to collect respiratory airflow signals clinically at Shandong Provincial Hospital.The dataset is constructed by the respiratory airflow signals from the left and right nasal cavities of each child,three experienced clinical physicians label samples based on the patient’s symptoms,medical history,nasal endoscopy or imaging,and other clinical information.A total of 128 children are recruited for this study,children with nasal structural deformities or lung diseases were excluded,a total of 74 children are enrolled in this study,and Dataset includes 148 segments of respiratory airflow signals.Based on the patient’s symptoms and the degree of adenoid hypertrophy evaluate by nasal endoscopy or imaging,53 children are included in the adenoid hypertrophy group,and 21 children are included in the healthy group.(2)Detection of adenoid hypertrophy in children based on respiratory shape.Five categories of features are extracted from the respiratory airflow signal,including Flattening,Scooping,Asymmetry,Timing and volume ratio measures,and Fluttering.Features are selected by Lasso,ANOVA,and the sequential forward feature selection algorithm based on linear discriminant analysis.The support vector machine with a linear kernel is used as the classifier to evaluate the single nasal respiratory airflow signal of the unilateral nasal cavity.Finally,the classification results of the bilateral nasal cavity are fused by OR operation to detect adenoid hypertrophy in children.The classification accuracy of the feature set selected by the sequential forward feature selection algorithm based on linear discriminant analysis is 86.93%.Meanwhile,by comparing three feature sets,this study finds that the dwell time of inspiratory airflow in children with adenoid hypertrophy is shorter than that in healthy children.(3)Detection of adenoid hypertrophy in children based on Respirdynamicsgram.Respirdynamicsgram(RDG)is generated by visualizing inherent respiratory dynamics underlying the nasal airflow modeled by dynamic learning.Due to adenoid hypertrophy,the airflow turbulence is increased,which results in the disorder of change in nasal airflow.Thus,the morphology of RDG remarkably differs between children with adenoidal hypertrophy(irregular shapes)and healthy children(regular butterfly-like shapes).Three features are utilized to represent the morphology of RDG,including temporal heterogeneity,spatial heterogeneity,and wing shape.The features for the morphology of RDG in the bilateral nasal cavity are fused by feature fusion based on exponential mapping,which detects and diagnoses adenoid hypertrophy in children.The fused features are fed into the linear support vector machine for the adenoid hypertrophy classification task.The mean accuracy over 100 resample folds is 89.14%.In summary,this thesis establishes the respiratory airflow signal database and develops two methods for the auxiliary diagnosis and detection of adenoid hypertrophy.Additionally,this thesis investigates the impact of adenoid hypertrophy on respiratory airflow from both the airflow shape and respiratory dynamics,which provides a new research idea and direction for aerodynamic analysis based on upper respiratory airflow.