Texture Analysis for Classification of Endometrial Tissue in Gray Scale Transvaginal Ultrasonography

Abstract

Computer-aided classification of benign and malignant endometrial tissue, as depicted in 2D gray scale transvaginal ultrasonography (TVS), was attempted by computing texture-based features. 65 TVS endometrial images were collected (15 malignant, 50 benign) and processed with a wavelet based enhancement technique. Two regions of interest (ROIs) were identified (endometrium, endometrium margin) on each processed image. Thirty-two textural features were extracted from each ROI employing first and second order statistics texture analysis algorithms. Textural feature-based models were generated for differentiating benign from malignant endometrial tissue employing stepwise logistic regression analysis. Models' performance was evaluated by means of receiver operating characteristics (ROC) analysis. The best benign versus malignant classification was obtained from the model combining three textural features from endometrium and four textural features from endometrium margin, with corresponding area under ROC curve (Az) 0.956.