Discrimination of invasive ductal and lobular carcinoma of the breast based on the combination of enhanced Legendre polynomial, kinetic features and deep learning features

Abstract

The fifth most common cause of cancer-related deaths among women worldwide is breast cancer, which is the most common cancer among women globally. Early detection of breast cancer through regular screenings and awareness of symptoms can lead to better prognosis and more effective treatment options. Breast cancer is a diverse disease that comes in more than 20 varieties. It is generally divided into two categories based on histology: in-situ carcinoma and invasive (infiltrating) carcinoma. These categories are further divided into four subcategories based on the location of the tumor’s origin: invasive ductal carcinoma (IDC), invasive lobular carcinoma (ILC), lobular carcinoma in situ (LCIS), and ductal carcinoma in situ (DCIS). The aim of this study is to use medical image processing and machine learning to accurately diagnose invasive lobular carcinoma (ILC) and invasive and lobular carcinoma (IDC) of the breast. This study develops a novel hybrid feature extraction model to improve the diagnosis accuracy of breast cancer. Deep learning, kinetic, and enhanced Legendre polynomial features have been combined to create a hybrid feature model that enhances IDC and ILC discrimination. The proposed model consists of four stages: data collection, preprocessing, feature extraction, and classification. The publicly available DCE-MRI dataset was used in this study, and the results showed classification accuracy of 97.99% in combined post-contrast-1 model. Overall, the results demonstrated the benefits of the hybrid feature extraction model and the fact that this study is non-invasive, uses only medical image processing, and does not require biopsies to enhance treatments.