Texture-based detection of lung pathology in chest radiographs using local binary patterns

This paper presents a method that employs texture-based feature extraction and Support Vector Machines (SVM) to classify chest abnormal radiograph patterns namely pleural effusion, pnuemothorax, cardiomegaly and hyperaeration. A similar previous attempt prototyped the classification system that achieved 97% and 87.5% accuracy for pleural effusion and pneumothorax using histogram values, while attaining 70% and 73.33% for cardiomegaly and hyperaeration using image processing schemes. In this work, we aimed to increase the performance in classifying the said lung patterns, specifically for cardiomegaly and hyperaeration. Using texture-based features, the developed system was able to achieve accuracies of 96% and 99% with sensitivities of 97% and 100% for the cardiomegaly and hyperaeration cases, respectively.