Diagnosis of COVID-19 on chest X-ray (CXR) images using CNN with transfer learning and integrated stacking ensemble learning

• Main Authors: Low, Wai Sing, Chow, Li Sze, Solihin, Mahmud Iwan, Handayani, Dini Oktarina Dwi
• Format: Proceeding Paper
• Language: English; English
• Published: Springer 2024
• Subjects: QA76 Computer software
• Online Access: http://irep.iium.edu.my/114201/1/114201_Diagnosis%20of%20COVID-19%20on%20chest%20X-ray%20%28CXR%29%20images.pdf; http://irep.iium.edu.my/114201/2/114201_Diagnosis%20of%20COVID-19%20on%20chest%20X-ray%20%28CXR%29%20images_SCOPUS.pdf; http://irep.iium.edu.my/114201/; https://link.springer.com/chapter/10.1007/978-981-99-8819-8_1
• Institution: Universiti Islam Antarabangsa Malaysia

COVID-19 caused a pandemic outbreak, resulting in many deaths and severe economic damage since 2019. Hence, the diagnosis of COVID-19 has be-come one of the major fields of research. Although RT-PCR has excellent relia-bility and precision, it is time-consuming and laborious. Therefore, the chest X-ray was used as an alternative and reliable diagnostic tool for COVID-19. How-ever, it requires a radiologist to analyze the X-ray images, which is limited by the availability of experts and time. Henceforth, many researchers deployed auto-mated computer-aided diagnosis with deep learning neural networks to speed up the diagnosis of COVID-19 with high accuracy and reproducibility. This study applied six state-of-art convolutional neural networks (DenseNet201, Mo-bileNetV2, ResNet101V2, VGG16, InceptionNetV3, and Xception) with transfer learning. An integrated stacking ensemble method was used to concatenate DenseNet201, MobileNetV2, VGG16, and Xception to produce a robust and ac-curate diagnostic model for COVID-19. The proposed ensembled CNN model in this study produced a test accuracy of 0.9725, sensitivity of 0.9749, and F1-score of 0.9724.