Early fusion techniques for enhanced pneumonia detection in chest X-ray analysis

Pneumonia, a significant global health concern, necessitates accurate diagnostic tools to enable timely and effective intervention. This study applies Early Fusion techniques to enhance pneumonia detection in chest X-ray analysis by integrating feature extraction methods—Principal Component Analysis (PCA), Histogram of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), and Local Binary Pattern (LBP). The fused features were used to train multiple machine learning models, including K-Nearest Neighbors (KNN), Logistic Regression, Random Forest, Neural Networks, and Support Vector Machines (SVM). Utilizing a dataset of 5,937 chest X-ray images, the SVM model achieved the highest accuracy (94.00%) and area under the curve (AUC) score (93.95%), demonstrating superior diagnostic performance compared to other models. The Early Fusion approach effectively captured complementary patterns in X-ray images, addressing challenges such as noise and subtle variability. These findings highlight the potential of fusion-based frameworks to improve diagnostic accuracy and reliability, supporting radiologists with automated tools for clinical decision-making. Future research will focus on validating the approach on larger, diverse datasets and improving the interpretability of fusion models while optimizing resource requirements to enable scalable deployment in real-world healthcare environments.