Attention-Driven Convolutional Neural Network for Accurate COVID-19 Diagnosis in Medical Imaging with GradCAM-Based Interpretability

The quick emergence of the COVID-19 (C19) pandemic has underscored the essential demand for precise and effective diagnostic instruments in medicine. Chest X-rays, a widely available and non-invasive imaging modality, have proven invaluable for diagnosing respiratory conditions, including C19. However, existing models often lack the interpretability and robustness required for clinical applications. This work addresses these gaps by proposing a custom CNN architecture augmented with an attention mechanism and dynamic learning rate adjustments. This hybrid model uses gradient-weighted class activation mapping (GradCAM) to make classification more accurate and predictions easier to understand. Our extensive data augmentation pipeline further ensures that the model is robust to real-world variations in medical imaging data. We trained and tested the proposed model on a carefully selected dataset. It did very well, with an accuracy of 99.09% and an area under the receiver operating characteristic curve of 1.00, beating out current top-of-the-line models. These results demonstrate the model's potential for reliable deployment in clinical settings, where interpretability and acc are paramount. Future research will investigate incorporating supplementary imaging modalities and real-time applications to augment the model's effectiveness in diagnosing respiratory illnesses.