Robust Texture Features For Emphysema Classification In CT Images
Haipeng Li, Ramakrishnan Mukundan · 2020
In this paper, we propose a novel feature extraction method based on local quinary patterns (LQP), multifractal features and intensity histograms for classifying emphysema into three subtypes in computed tomography images. Compared to local binary patterns, LQP method computes more image local patterns to represent texture features. Multifractal features which enhancing local textures are combined with other features to constitute a feature vector for this classification task. An autoencoder network and principal components analysis are used to reduce the dimensionality of the feature vector before using an SVM classifier. The proposed method is tested on an emphysema database containing 168 annotated regions of interest of three different subtypes. The experimental results demonstrate that our method outperforms most other state-of-the-art approaches with the best classification accuracy of 92.3% using the least dimensionality (15) of the feature vector.