| The COVID-19 pandemic has become the most notable social event,affecting all aspects of people’s lives.Mass and long-term nucleic acid testing have put tremendous pressure on the medical system and consumed a significant amount of resources.Therefore,it is necessary to develop fully automated COVID-19 testing methods.Additionally,centralized nucleic acid testing carries the risk of cluster infections and has a longer testing cycle.Therefore,fully automated COVID-19 testing is essential.Clinical studies have found that ground glass opacities and pulmonary nodules often appear in the lung CT images of COVID-19 patients.Clinical observations and studies have found that ground-glass opacities or pulmonary consolidation often appear in the lung CT images of COVID-19 patients.Based on this fact,many scholars have extensively studied COVID-19 detection methods based on lung imaging.This paper conducts research from multiple perspectives,including feature extraction,feature fusion,comprehensive diagnosis of chest CT scans of COVID-19 patients,and rapid development of an automated COVID-19 detection system.A COVID-19 detection system that can be rapidly developed and implemented in a short period of time is proposed.The main research content of this article includes:(1)In order to overcome the lack of pixel-level COVID-19 lung segmentation data,this article did not use an additional segmentation network to extract lesion locations.Instead,a lightweight attention mechanism was added to the convolutional neural network to ensure accurate localization of the lesion areas without significantly increasing the network scale.Additionally,a feature extraction network with an encoder-decoder structure was constructed to achieve multiscale feature fusion,effectively improving network performance.This article proposes a comprehensive diagnosis method based on the correlation between consecutive CT images of the lungs,alleviating the problem of "overly sensitive" convolutional neural networks.(2)This paper proposes a strategy based on federated learning to enable multiple medical institutions to jointly model,and success was achieved in simulation experiments.This approach greatly reduces the time required for data collection and facilitates the construction of a "data sharing alliance" to break the "data island" dilemma without compromising patient privacy.(3)This paper designs and implements a web-based prototype of the COVID-19 detection system.The deep neural network model is implemented using the Pytorch framework,the system backend is built using Springboot,and the pages of the system are built using Vue2.0 frameworks.My SQL database is used to store information.The system achieves basic functions such as user login,patient basic information entry,CT image upload and COVID-19 assisted diagnosis,patient case inquiry,and personnel management. |
