Analyzing the Texture of Nakagami Parametric Images for Classification of Breast Cancer

Abstract

In this paper we analyze the capability of texture features extracted from Nakagami parametric images for the classification of breast cancer. Nakagami parametric maps were generated from ultrasound envelope images using a sliding window of length 0.75mm and 0.0385mm increment (95% overlap). Next, Gray Level Co-occurrence Matrix (GLCM) techniques were applied to the parametric maps in order to extract texture features. These texture features were utilized for the classification of breast lesions. An Area under the Receiver Operating Characteristics curve (AUC) of 0.90 and a sensitivity of 88.5% was obtained, suggesting that texture features derived from Nakagami parametric images have the potential to play an important role in the early diagnosis of breast cancer.