Improved Classification of Ultrasound Breast Cancer RF Data with Nakagami and Derived Nakagami Parameters Using Advanced Machine Learning Algorithm

Abstract

Ultrasound imaging is one of the medical imaging modalities that has become popular among the researchers for breast cancer diagnosis because of its radiation free and non-invasive nature compared to other screening procedures like X-ray Mammography, CT-scan etc. In ultrasound studies, various quantitative analysis has been shown to be capable of breast cancer diagnosis. This research utilizes the statistical parameters from the well-known Nakagami distribution and investigates their potential in improvement of noninvasive semi-automated identification of breast cancer. The dataset used in the study has 130 biopsy-proven patients consisting of 104 benign and 26 malignant cases with traced lesion boundaries. In this study, seven types of Nakagami and derived Nakagami images have been generated for each patient from the basic and derived parameters of Nakagami distribution. To determine the suitable window size for image generation, an empirical analysis has been conducted using three window sizes of width 0.2 mm and lengths 0.1875 mm, 0.45 mm and 0.75mm. The images were analyzed quantitatively for feature extraction, followed by feature selection using the machine learning algorithm- Recursive Feature Elimination with Cross Validation. A combination of six features was found for window size 0.75mm which gives a classification accuracy of 92.3% and area under the ROC curve score of 0.95 using Linear Support Vector Machine classifier.