An Evaluation of Effectiveness of a Texture Feature Based Computerized Diagnostic Model in Classifying the Ovarian Cyst as Benign and Malignant from Static 2D B-Mode Ultrasound Images

Abstract

Objective: To develop a computerized diagnostic model to characterize the ovarian cyst at its early stage in order to avoid unnecessary biopsy and patient anxiety. Background: The main cause of mortality and infertility in women is ovarian cancer. It is very difficult to diagnose ovarian cancer using ultrasonography as benign and malignant ovarian masses or cysts exhibit similar characteristics. Early prediction and characterization of ovarian masses will reduce the unwanted growth of the ovarian mass. Materials and Methods: Transvaginal 2D B mode ovarian mass ultrasound images were preprocessed initially to enhance the image quality. And then, the region of interest (ROI) in this case ovarian cyst was segmented. Finally, Local Binary Pattern (LBP) textural features were extracted. A Support Vector Machine was trained to classify the ovarian cyst or mass as benign or malignant. Results: The performance of the SVM improved with an average accuracy of 92% when the textural features were extracted from the Original Gray Value-based LBP (OGV-LBP) image than the histogram- based LBP. Conclusion: The SVM can classify the transvaginal 2D B mode ovarian cyst ultrasound images into benign and malignant effectively when the textural features from the original gray value-based LBP extracted were considered.