An Efficient Learning based Algorithm for Lung Boundary Detection for Chest X-ray Images

Savitha S. K, Aprameya K. S, Alwyn R. Pais

Abstract We present an algorithm for the automatic delineation of lung boundary fields in chest radiographs. An attempt has been made in this paper to develop a novel mechanism for the detection of lung boundaries. We proposed a new Constrained Active Shape Modeling (CASM) algorithm where ASM algorithm is constrained by the Support Vector Machine (SVM) model parameters during the lung boundary detection in chest x-ray images. Initially the images are preprocessed and subjected to rough segmentation module which generates roughly segmented lung boundaries and then lung boundary specific geometric parameters are computed. In the final stage, later this rough segmentation boundary is used for optimal convergence. Simultaneously we generate a subspace of lung boundary landmark distribution using SVM for all the images in the database. In the next step the detected lung boundary landmark distribution for all the images is subjected to training using ASM algorithm. Finally the rough segmentation lung boundary converges to the optimal lung boundary during CASM application. The proposed method is tested on digital chest image database comprising of various patient populations. The performance evaluation of the proposed method is quantified using degree of overlap between the segmentation result and ground truth result. We also used average contour distance measure to assess the accuracy of lung boundary detection. Both the measures yield promising results in comparison with existing methods. Keywords â€“ ASM, SVM, level sets, edge gradient, X-ray

Title:	An Efficient Learning based Algorithm for Lung Boundary Detection for Chest X-ray Images
Author Name:	Savitha S. K, Aprameya K. S, Alwyn R. Pais
Abstract:	Abstract We present an algorithm for the automatic delineation of lung boundary fields in chest radiographs. An attempt has been made in this paper to develop a novel mechanism for the detection of lung boundaries. We proposed a new Constrained Active Shape Modeling (CASM) algorithm where ASM algorithm is constrained by the Support Vector Machine (SVM) model parameters during the lung boundary detection in chest x-ray images. Initially the images are preprocessed and subjected to rough segmentation module which generates roughly segmented lung boundaries and then lung boundary specific geometric parameters are computed. In the final stage, later this rough segmentation boundary is used for optimal convergence. Simultaneously we generate a subspace of lung boundary landmark distribution using SVM for all the images in the database. In the next step the detected lung boundary landmark distribution for all the images is subjected to training using ASM algorithm. Finally the rough segmentation lung boundary converges to the optimal lung boundary during CASM application. The proposed method is tested on digital chest image database comprising of various patient populations. The performance evaluation of the proposed method is quantified using degree of overlap between the segmentation result and ground truth result. We also used average contour distance measure to assess the accuracy of lung boundary detection. Both the measures yield promising results in comparison with existing methods. Keywords â€“ ASM, SVM, level sets, edge gradient, X-ray
Cite this article:	Savitha S. K, Aprameya K. S, Alwyn R. Pais , " An Efficient Learning based Algorithm for Lung Boundary Detection for Chest X-ray Images " , International Journal of Emerging Trends & Technology in Computer Science (IJETTCS), Volume 3, Issue 4, July - August 2014 , pp. 291-298 , ISSN 2278-6856.
Full Text [PDF] Home