Ensemble Learning-based Breast Tumor Classification Using Ultrasound Images

Breast cancer is one of the leading causes of cancer death in women, making early identification crucial for appropriate and timely treatment. The breast tumors are typically classified into two categories: benign and malignant. In clinical applications, ultrasound imaging is one of the most commonly used non-invasive ways for breast tumor imaging. In this paper, we propose an ensemble learning-based method for malignant tumor identification, assisting clinicians in accurate breast cancer diagnosis. Firstly, multiple tumor detection networks are trained, and then the detected bounding boxes are grouped to merge the detected regions. Furthermore, a novel voting operation is designed to obtain the tumor region and the classification result. In this study, 7311 images collected from various devices are used for training and testing. Experimental results show that our proposed ensemble learning method can achieve more accurate cancer identification results compared with a single network.