Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Focal Neural Network for the Imbalanced COVID-19 Imaging Data
1School of Electronic and Communication Engineering, Shenzhen Polytechnic University, China.
2College of Mathematics and Statistics, Shenzhen University, China.
ABSTRACT
Coronavirus disease 2019 (COVID-19) has become a worldwide epidemic for almost two years. Chest X-ray imaging is one of the best available methods for early diagnosing of the COVID-19. In this paper, a focal neural network has been designed for the imbalanced COVID-19 X-ray imaging data. A modified CheXNet has been used to extract the deep features. A focal loss function has been introduced in our network, which helps to down-weight the major negative samples in the imbalanced COVID-19 imaging data and focus on the correct classification of rare positive samples. Considering the sparse and nonlinear characteristics of deep features, we use XGBoost algorithm to improve the performance of classifiers. We have also set up the biggest COVID-19 chest X-ray imaging dataset, which consists of more than 44000 chest X-ray images and includes more than 8000 positive samples of COVID-19 patients. Extensive experiments have verified the state-of-the-art performance of our model.
Keywords: Neural network, COVID-19, Imbalanced data, Focal loss.
Download PDF
1School of Electronic and Communication Engineering, Shenzhen Polytechnic University, China.
2College of Mathematics and Statistics, Shenzhen University, China.
ABSTRACT
Coronavirus disease 2019 (COVID-19) has become a worldwide epidemic for almost two years. Chest X-ray imaging is one of the best available methods for early diagnosing of the COVID-19. In this paper, a focal neural network has been designed for the imbalanced COVID-19 X-ray imaging data. A modified CheXNet has been used to extract the deep features. A focal loss function has been introduced in our network, which helps to down-weight the major negative samples in the imbalanced COVID-19 imaging data and focus on the correct classification of rare positive samples. Considering the sparse and nonlinear characteristics of deep features, we use XGBoost algorithm to improve the performance of classifiers. We have also set up the biggest COVID-19 chest X-ray imaging dataset, which consists of more than 44000 chest X-ray images and includes more than 8000 positive samples of COVID-19 patients. Extensive experiments have verified the state-of-the-art performance of our model.
Keywords: Neural network, COVID-19, Imbalanced data, Focal loss.
Download PDF