Enhanced ResNet-50 with Multi-Feature Fusion for Robust Detection of Pneumonia in Chest X-Ray Images

Background/Objectives: Pneumonia is a critical lung infection that demands timely and precise diagnosis, particularly during the evaluation of chest X-ray images. Deep learning is widely used for pneumonia detection but faces challenges such as poor denoising, limited feature diversity, low interpretability, and class imbalance issues. This study aims to develop an optimized ResNet-50 based framework for accurate pneumonia detection. Methods: The proposed approach integrates Multiscale Curvelet Filtering with Directional Denoising (MCF-DD) as a preprocessing step to suppress noise while preserving diagnostic details. Multi-feature fusion is performed by combining deep features extracted from ResNet-50 with handcrafted texture descriptors such as Local Binary Patterns (LBPs), leveraging both semantic and structural information. Precision attention mechanisms are incorporated to enhance interpretability by highlighting diagnostically relevant regions. Results: Validation on the Kaggle chest radiograph dataset demonstrates that the proposed model achieves higher accuracy, sensitivity, specificity, and other performance metrics compared to existing methods. The inclusion of MCF-DD preprocessing, multi-feature fusion, and precision attention contributes to improved robustness and diagnostic reliability. Conclusions: The optimized ResNet-50 framework, enhanced by noise suppression, multi-feature fusion, and attention mechanisms, offers a more accurate and interpretable solution for pneumonia detection from chest X-ray images, addressing key challenges in existing deep learning approaches.