Diagnosis Of Adult OSA Based On Craniofacial Lateral Film

Obstructive sleep apnea syndrome（OSAS）is a disease with repeated partial or total failure of the upper respiratory tract.Most obstructive sleep apnea is related to different degrees of skull and bone changes.Mild diseases will affect the quality of patients’ night sleep,and severe cases will lead to other types of diseases.At present,the main detection methods of obstructive sleep apnea are multi-channel sleep monitoring and traditional machine learning.Due to the defects of high cost and low efficiency in multi-channel sleep monitoring and diagnosis,and the traditional machine learning heavily relies on manual labeling data.Therefore,this paper takes the low-cost craniofacial lateral X-ray film as the data set,expand the sample data through Deep Convolution Generative Adversarial Networks data enhancement algorithm,and proposes a new Res Net-OSA deep learning diagnosis method to diagnose obstructive sleep apnea and provide discrimination probability.And the adult OSA diagnosis system based on this method is realized.This study can effectively reduce the cost of diagnosis and reduce the workload of doctors.The main work of this subject includes the following two aspects:（1）Based on the key point detection model,the lateral craniofacial X-ray films are located and detected by cascade regression,and then the key points in the remaining data sets are predicted by using the framework to obtain the key point position information corresponding to the lateral craniofacial films of all subjects.The model is iteratively optimized by semi supervised learning,and finally classified by using the classification model,Finally,it is found that this diagnostic method is useful for the identification of obstructive sleep apnea syndrome,which also provides a evaluation reference for the follow-up deep learning model.（2）A deep learning model called Res Net-OSA is proposed.The model mainly uses the feature pyramid network principle to expand the receptive field area of convolution kernel,and adds dropout layer to improve the classification effect.Finally,the data set enhanced by Deep Convolution Generative Adversarial Networks is used for grouping training,and the classification effect of the model on obstructive sleep apnea in different regions is detected.The classification performance of the model,the key point detection model and the existing deep learning model is compared and analyzed.The results show that the model has the best performance in the classification of obstructive sleep apnea,and the classification accuracy is91.83%.In this paper,deep learning technology and key point detection technology are applied to the classification of lateral craniofacial images,and the diagnosis system of obstructive sleep apnea is realized.The results show that the model proposed in this paper can achieve good results,which can help doctors improve the accuracy and efficiency of obstructive sleep apnea judgment,and has strong practical significance.