CORE-MONET: A Multi-Stage Ensemble Deep Learning Framework for Automated COVID-19 Detection and Explainable Lung Infection Severity Assessment from Chest X-rays

COVID-19 is a contagious disease caused by SARS-CoV-2 infection. Chest X-ray (CXR) imaging provides a faster alter-native for COVID-19 screening. However, manual interpretation is subjective and prone to errors. Deep learning models can automate COVID-19 detection with high accuracy. This study proposes an ensemble model combining ResNet50 and MobileNetV2. ResNet50 extracts deep hierarchical features for precise classification. MobileNetV2 enhances effi-ciency while maintaining strong performance. A segmentation step isolates lung regions to improve feature extraction. Grad-CAM enhances explainability by highlighting infection-prone areas. The dataset includes CXR images from the COVID-19 Radiography Database. Images were split 80:10:10 for training, validation, and testing with stratified sam-pling. The ensemble model integrates feature fusion for improved classification. The model was optimized using Adam optimizer (learning rate: 0.0001, β1=0.9, β2=0.999) for 10 epochs with early stopping (patience=3, monitoring valida-tion loss). Performance evaluation considers accuracy, sensitivity, specificity. Using stratified fivefold cross-validation the model achieved 97.55% accuracy, precision of 97.2%, F1-score of 97.4% and AUC-ROC of 0.989 in differentiating
COVID-19 from normal cases. Grad-CAM heatmaps confirm the model's focus on infection regions. Infection severity
is quantified using lung segmentation and Gradient-weighted Class Activation Mapping (Grad-CAM) overlays. Higher
severity scores correspond to severe lung involvement in patients. The proposed method enhances COVID-19 detection
and interpretability.