TY - GEN
T1 - Automatic Detection of COVID-19 Using Ensemble Transfer Learning Based on Lung CT Scans
AU - Pillay, Ricardo
AU - Viriri, Serestina
AU - Heymann, Reolyn
N1 - Publisher Copyright:
© 2023, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering.
PY - 2023
Y1 - 2023
N2 - In order to curb the rapid spread of COVID-19, early and accurate detection is required. Computer Tomography (CT) scans of the lungs can be utilized for accurate COVID-19 detection because these medical images highlight COVID-19 infection with high sensitivity. Transfer learning was implemented on six state-of-the-art Convolutional Neural Networks (CNNs). From these six CNNs, the three with the highest accuracies (based on empirical experiments) were selected and used as base learners to produce hard voting and soft voting ensemble classifiers. These three CNNs were identified as Vgg16, EfficientNetB0 and EfficientNetB5. This study concludes that the soft voting ensemble classifier, with base learners Vgg16 and EfficientNetB5, outperformed all other ensemble classifiers with different base learners and individual models that were investigated. The proposed classifier achieved a new state-of-the-art accuracy on the SARS-CoV-2 dataset. The accuracy obtained from this framework was 98.13%, the recall was 98.94%, the precision was 97.40%, the specificity was 97.30% and the F1 score was 98.16%.
AB - In order to curb the rapid spread of COVID-19, early and accurate detection is required. Computer Tomography (CT) scans of the lungs can be utilized for accurate COVID-19 detection because these medical images highlight COVID-19 infection with high sensitivity. Transfer learning was implemented on six state-of-the-art Convolutional Neural Networks (CNNs). From these six CNNs, the three with the highest accuracies (based on empirical experiments) were selected and used as base learners to produce hard voting and soft voting ensemble classifiers. These three CNNs were identified as Vgg16, EfficientNetB0 and EfficientNetB5. This study concludes that the soft voting ensemble classifier, with base learners Vgg16 and EfficientNetB5, outperformed all other ensemble classifiers with different base learners and individual models that were investigated. The proposed classifier achieved a new state-of-the-art accuracy on the SARS-CoV-2 dataset. The accuracy obtained from this framework was 98.13%, the recall was 98.94%, the precision was 97.40%, the specificity was 97.30% and the F1 score was 98.16%.
KW - Convolutional neural network
KW - Deep learning
KW - Hard voting
KW - Soft voting
KW - Transfer learning
UR - http://www.scopus.com/inward/record.url?scp=85149656467&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-25271-6_12
DO - 10.1007/978-3-031-25271-6_12
M3 - Conference contribution
AN - SCOPUS:85149656467
SN - 9783031252709
T3 - Lecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering, LNICST
SP - 181
EP - 204
BT - Pan-African Artificial Intelligence and Smart Systems - Second EAI International Conference, PAAISS 2022, Proceedings
A2 - Ngatched Nkouatchah, Telex Magloire
A2 - Woungang, Isaac
A2 - Tapamo, Jules-Raymond
A2 - Viriri, Serestina
PB - Springer Science and Business Media Deutschland GmbH
T2 - 2nd International Conference on Pan-African Intelligence and Smart Systems, PAAISS 2022
Y2 - 2 November 2022 through 4 November 2022
ER -