Gradient contouring and texture modelling based CAD system for improved TB classification

Abstract

In this work, fully automated software design is developed for TB recognition system which includes deformable gradient based active contour level set model for isolating the lung region from input chest x-ray images. In general, segmenting the lung region from CXR images computationally intensive task due its complex analyzes, dynamic morphological variations among different classes and boundary discontinuities etc. In particular, as compared to all other abnormality analyzes in CXR images TB detection required most appropriate ROI segmentation irrespective among image non linearity’s. The proposed method considers ACM modeling by eliminating discontinuous boundary conditions. Here in this work gradient information at the lung boundaries active contour model is derived. This reduces the computational cost and increases accuracy as a consequence of selecting optimized global threshold limit and gradient features in all possible ways. This framework also includes unified texture classification model to enrich the texture content from the ROI segmented lungs regions from CXR imaging. To meet this requirement, orientation driven texture classification is done which retain texture information’s from all the possible coordinate angles to accomplish comprehensive texture retention. More generally, the intrinsic relationship between texture classification and feature set modeling has been explicitly analyzed and presented. Moreover, state-of-the-art feature extraction and selection (CBH-FS) has been introduced and embedded into the framework to form a complete, automated Tuberculosis detection system. The whole system was successfully evaluated by several benchmark datasets and it was shown that the algorithms mentioned earlier efficiently detect TB affected lung images. Finally supervised support vector machine (SVM) based Artificial Intelligence (AI) learning model is used to validate the false detection rate of fully automated TB classification CAD software system.