Fusion Framework for Morphological and Multispectral Textural Features for Identification of Endometrial Tuberculosis

Abstract

Endometrial Tuberculosis (ETB) is primarily diagnosed in infertile females as a fallout of Female Genital Tuberculosis (FGTB). An effective and fast computational method to diagnose ETB from Transvaginal ultrasound (TVUS) images is of great importance to the community. The objective of this paper is to obtain an optimal subset of features for an effective and discriminative analysis of TVUS images for identifying ETB. The TVUS images from different medical centers in India have been collected under expert supervision from female patients. Texture and Morphological features effectively capture the observations made by the experts for identifying the problem in hand. Therefore a fusion framework model is proposed where the extracted image features are fused and an optimal subset of features is obtained for identification. Multiresolution transformation of ill-defined TVUS images highlights the directional, multi- scale spectral textural features. Therefore, to obtain discriminatory textural features, images are transformed using Non-Subsampled Contourlet Transformation (NSCT) before feature extraction. Experimental results of the fusion model for classification show significant improvements and prove to be more efficient. The proposed methodology records an F-score of 0.845 with a sensitivity score of 0.818 for the dataset available. A feature reduction of 64.5% is attained for the classification of the dataset after feature selection.