International Conference on Computer and Knowledge Engineering

Home / 13th International Conference on Computer and Knowledge Engineering

Classification of benign and malignant tumors in Digital Breast Tomosynthesis images using Radiomic-based methods

Authors :

Farangis Sajadi moghadam¹ Saeid Rashidi²

1- Medical Sciences & Technologies Faculty, Science & Research Branch, Islamic Azad University, Te 2- Medical Sciences & Technologies Faculty, Science & Research Branch, Islamic Azad University, Te

Keywords :

Breast Cancer،Feature Extraction،Learning Algorithm،Radiomic،Tomosynthesis Images

Abstract :

Breast cancer arises from the uncontrolled proliferation of abnormal cells, leading to the formation of a mass in the breast tissue. Digital Breast Tomosynthesis (DBT), a three-dimensional imaging technology, has enhanced both screening and diagnostic outcomes. It provides supplementary information that mitigates the confounding effects of tissue overlap and enhances the detection, identification, and localization of abnormalities. The objective of this research is to classify the benign or malignant nature of masses in DBT images using Radiomic features. This analysis utilizes an open database from TCIA consisting of 224 lesion bounding boxes. To effectively extract relevant features, a two-dimensional central slice of the DBT image encompassing a significant anatomical portion of the breast tumor is utilized. During the pre-processing stage, the rescale intensity method is employed to enhance contrast and improve image quality. Subsequently, a binary mask is utilized to segment the breast tissue mass. Four categories of Radiomic features are then extracted. The study investigates the suitability of these features for benign-malignancy classification. Furthermore, the impact of feature selection, feature balancing, and feature normalization is explored in conjunction with eight different learning algorithms. With this setting, the best result of the evaluation metrics in terms of mean AUC, accuracy, sensitivity and specificity are equal to 88.56%, 88.67%, 77.12 % and 75.11% for Quadratic Discriminant Analysis (QDA), respectively.