Pneumonia and Tuberculosis Classification from Chest X-ray Images Through Composite Features and Ensemble Learning

This paper proposed a simple and effective lung diseases classification model from CXR images through machine learning (ML) algorithms. Unlike the existing methods which used complex deep learning algorithms, this model is very simple and focused on ensuring maximum discrimination through handcrafted features. Four different set of features are used to describe CXR image in four different orientations like texture, geometry, intensity and high frequencies. At classification, an ensemble learning strategy is proposed which comprised of two ML algorithm namely Support Vector Machine (SVM) and Extreme learning Machine (ELM). SVM differentiates the input CXR images into two classes namely normal and diseased. Next, ELM differentiates the diseases image into two classes namely Pneumonia and Tuberculosis. Additionally, preprocessing is applied to improve the CXR image quality in terms of contrast enhancement and noise removal. Further, to reduce the complexity, this work processed only a Region of Interest for feature extraction followed by classification. Towards ROI extraction, a dynamic threshold mechanism is applied which drives the threshold purely based on image morphological characteristics. Extensive experiments are carried out over the proposed model with varying simulation scenarios and the obtained results prove that the proposed model is superior to the state-of-the-art methods. It aims to emphasize the significance of early detection in lung diseases, the limitations of current methods, and the suggested alternatives. The average accuracy gained by proposed model is observed as 98% which is larger than all the existing approach.