Early Diagnosis of Lung Diseases with Deep Learning Using EfficientNetV2-S Architecture

Lung disease diagnosis through deep learning has shown promising advancements, yet existing models often face challenges in accuracy, computational efficiency, and interpretability. This study proposes an enhanced deep learning framework leveraging EfficientNetV2 S, a state-of-the-art convolutional neural network, for automated lung disease classification. The model is optimized to achieve higher accuracy, reduced computational cost, and improved interpretability compared to traditional architectures such as ResNet and DenseNet. The proposed method undergoes rigorous evaluation on a bench- mark medical imaging dataset, where it achieves an accuracy of 96.5%, surpassing prior models. Additionally, Grad-CAM visualizations demonstrate the model's ability to effectively localize diseased regions, offering clinicians interpretability in AI- driven decision-making. To validate the statistical significance of these improvements, paired t-tests and confidence interval analysis are conducted, confirming a pvalue <0.01, signifying a significant performance enhancement. The model achieves a 9.2% improvement in classification accuracy while reducing computational overhead by 30%, making it more efficient for real-time deployment. Furthermore, the integration of Grad- CAM-based attention mechanisms enhances model transparency, providing a visual justification for AI-driven decisions. This research bridges the gap between AIdriven diagnostics and real- world medical applicability, offering a scalable and interpretable solution for automated lung disease detection.