Gastroscopy Image Analysis Method Based Onintegrated Learning And Design Of Mobile Medical Platform

Endoscopy has always been one of the important means for doctors to diagnose patients' conditions in the medical system.Endoscopic imaging technology has been developing continuously,but how to analyze endoscopic imaging precisely and diagnose follow-up diseases rapidly are extremely great challenges for doctors.Gastroscopy,as a common type of endoscopy,has a higher requirement for the ability of examiners to identify lesions and judge their condition because of the high similarity of images and the concealment of lesions.Therefore,how to effectively assist doctors in the analysis and diagnosis of gastroscope imaging through technical means,reduce the probability of misdiagnosis and missed diagnosis is an urgent problem to be solved.In addition,with the rapid development of mobile Internet,smart mobile terminals have become more and more popular.Accompanied by this,the development of mHealth(mobile health)is very rapid,and more and more attention has been paid to mHealth by medical,electronic technology and other related fields.A large number of mobile medical APPs with different functions and different application scenarios have been born,which greatly meet the needs of users for convenient medical treatment.Based on the above mentioned scenarios,the thesis implemented a method based on integrated learning and convolution neural network to analyze gastroscopic image,which can help doctors quickly find and determine the location of lesions,improve the efficiency of doctors,and avoid possible missed diagnosis and misdiagnosis by inexperienced doctors.This method is based on AdaBoost's integrated learning method,which combined four subclassifiers composed of AlexNet,GoogLeNet,VGGNet and ResNet to construct a total classifier by assigning different weight parameters.The final prediction results are output by accumulating weights.Through experiments and comparison with the existing methods,we found that this method effectively improves the accuracy of traditional gastroscopy,and it had better performance than traditional methods in the following four indicators,sensitivity,specificity,missed diagnosis rate and misdiagnosis rate,which can be applied in practice.And also,the thesis designed an mHealth platform—— "Haola Doctor" mHealth platform,which focuses on providing live operation,remote consultation and video conference.The gastroscopy analysis function as mentioned above will also be integrated into the "Haola Doctor" medical service platform in the future,so that the platform's functionality can be enhanced and the application scope of the platform will be greatly extended.The platform adopts C/S architecture and MVC design model,which can be divided into three parts: server side,client side and Web side.The client side can also be divided into three parts: iOS,Android client APP and Android push-flow APP which is Android-only pushed for authorized users.The thesis will systematically elaborate the whole process from the generation of requirements to the final development and successful release of the two APPs on Android side and server side.The technologies involved include the development of Android based on xUtils framework and the construction of background Redis and MySQL database based on JFinal.The platform is developed in depth based on hardware,which makes the hardware more adaptable and less functional redundancy.The function is more in line with the needs of users,and it's less difficult to start,more convenient to operate for users.