Fig 1.
Transfer learning example: DenseNet201 pre-trained feature extraction layers.
Table 1.
Comparison of the well-known CNN models used in deep transfer learning on the number of parameters.
Fig 2.
(A-D) COVID-19 positive cases and (E-H) COVID-19 negative cases.
Fig 3.
The proposed framework for COVID-19 detection.
Fig 4.
The proposed CNN model for feature extraction.
Table 2.
Testing scores of the proposed methods for COVID-19 classification.
Fig 5.
ROC AUC curve (a) the proposed CNN model+random forest classifier (b) the proposed CNN model+logistic regression classifier (c) tuned MobileNetV2 + random forest classifier.
Fig 6.
Precision-Recall curves (a) the proposed CNN model+random forest classifier (b) the proposed CNN model+logistic regression classifier (c) tuned MobileNetV2 + random forest classifier.
Fig 7.
Confusion matrices (a) the proposed CNN model+random forest classifier (b) the proposed CNN model+logistic regression classifier (c) tuned MobileNetV2 + random forest classifier.
Fig 8.
Evaluation metrics of tuned MobileNetV2 architecture (a) Training and test accuracy (b) Training and test loss.
Fig 9.
The proposed model’s accuracy with 95% confidence interval for 10 cross-validation folds.
Table 3.
Comparison of the proposed model and the state-of-the-art DL-based models for COVID-19 classification.