Research Article
COVID-19 Classification Algorithm Based on Privacy Preserving Federated Learning
@INPROCEEDINGS{10.1007/978-3-031-34586-9_13, author={Changqing Ji and Cheng Baoluo and Gao Zhiyong and Qin Jing and Wang Zumin}, title={COVID-19 Classification Algorithm Based on Privacy Preserving Federated Learning}, proceedings={Pervasive Computing Technologies for Healthcare. 16th EAI International Conference, PervasiveHealth 2022, Thessaloniki, Greece, December 12-14, 2022, Proceedings}, proceedings_a={PERVASIVEHEALTH}, year={2023}, month={6}, keywords={Federated learning COVID-19 Convolutional block attention module Resnet50 Privacy preserving}, doi={10.1007/978-3-031-34586-9_13} }- Changqing Ji
Cheng Baoluo
Gao Zhiyong
Qin Jing
Wang Zumin
Year: 2023
COVID-19 Classification Algorithm Based on Privacy Preserving Federated Learning
PERVASIVEHEALTH
Springer
DOI: 10.1007/978-3-031-34586-9_13
Abstract
In order to solve the problems of complex feature extraction, slow convergence of model and most of the deep learning based COVID-19 classification algorithms ignore the problem of “island” of medical data and security. We innovatively propose a COVID-19 X-ray images classification algorithm based on federated learning framework, which integrates hybrid attention mechanism and residual network. The algorithm uses hybrid attention mechanism to highlight high-resolution features with large channel and spatial information. The average training time is introduced to avoid the long-term non-convergence of the local model and accelerate the convergence of the global model. For the first time, we used the federated learning framework to conduct distributed training on COVID-19 detection, effectively addressing the data “islands” and data security issues in healthcare institutions. Experimental results show that the Accuracy, Precision, Sensitivity and Specific of the proposed algorithm for COVID-19 classification on datasets named’ COVID-19 Chest X-ray Database’ can reach 0.939, 0.921, 0.928 and 0.947, respectively. The convergence time of the global model is shortened by about 30 min. That improves the performance and training speed of the COVID-19 X-ray image classification model with privacy security.
