Fast and efficient lung disease classification using hierarchical one-against-all support vector machine and cost-sensitive feature selection

A hierarchical support vector machine (SVM) with class-specific feature selection was proposed to enhance the accuracy and reduce the classification time for differentiating diffuse interstitial lung disease in computer-aided quantification. This method allowed each binary classifier to use a class-specific quasi-optimal feature set, and a computational cost-sensitive group-feature selection criterion was applied for accelerating the classification time. Compared to one-against-all and one-against-one SVM methods with sequential forward selection, the proposed method reduced classification time by up to 57% and significantly improved overall accuracy (paired t-test, p<0.001). These results suggest potential for the proposed method in real-time and online image-based clinical applications. This study also had an impact on the development of AVIEW Lung Texture software.