Performance of support vector machines in pneumonia detection using chest x-ray images from Filipino cohorts
Studies have shown that computer-aided diagnosis (CAD) systems significantly improve the accuracy and speed of radiologic interpretations of chest x-ray images (CXR). In this study, we developed a machine learning-based CXR image classifier by optimizing support vector machine (SVM) to distinguish pneumonia from normal CXR images. The dataset used to develop the classifier model were generated through a retrospective clinical study conducted in the Philippines. We implemented feature extraction on the pre-processed CXR by considering four statistical feature sets: intensity histogram, invariant moments, Haralick features, and local binary pattern. Results show that the SVM classifier model has 88.33% accuracy, 86.57% precision, 90.63% sensitivity, 86.05% specificity, and 0.94 AUC in distinguishing pneumonia from normal CXR. These results illustrate the potential of a machine learning-based model to facilitate high sensitivity and automated screening of pneumonia.