Optimized Deep Learning Models Using Ensemble Learning for COVID-19 Detection on CT Scan Images

Abstract

Early identification of COVID-19 can facilitate the establishment of a swift medical response plan, thereby slowing the rapid dissemination of this deadly disease. Recent advancements in medical imaging technology, coupled with the successful application of deep learning to visual tasks, have driven numerous studies investigating early disease diagnosis through medical imaging. In particular, deep learning has been employed for COVID-19 diagnosis from CT scan images. This paper proposes an ensemble COVID detection model that integrates four models including GoogleNet, EfficientNet, Hybrid EfficientNet and DOLG, and DenseNet121. And an augmentation dataset for training/testing is utilized in this work. To comparing performance, we train four models separately with our ensemble model and evaluate their results. The results demonstrate that our ensemble model outperforms all individual models, obtaining a detection accuracy of 94 percents. Hence, our proposed ensemble model shows the potential of the ensemble method in improving the accuracy of COVID diagnosis.