Diffuse lung disease classification based on texture features and weighted extreme learning machine

Abstract

Diffuse lung diseases are a group of chronic disorders that affect the lungs. The highly prevalent lung patterns associated with diffuse lung diseases are emphysema, fibrosis, ground-glass opacity, and micro-nodules. For diffuse lung classification problem, TALISMAN (Texture Analysis of Lung ImageS for Medical diagnostic AssistaNce) is one of the widely studied dataset in the literature. It is observed in the dataset that there exists sample imbalance among different tissue patterns. To address the sample imbalance in the data weighted extreme learning machine classifier is employed in this work. To overcome the intra-class and inter-class variation among the diffuse lung patterns features are extracted using the modified intuitionistic local binary pattern along with Gabor filter bank and grey level co-occurrence matrix. These combined texture features are then used to train the weighted extreme learning machine to classify the diffuse lung patterns. The performance of the proposed approach is compared with the existing works in the literature. The comparison results indicate better performance of the proposed approach for diffuse lung classification with sample imbalance.