BREAST CANCER CLASSIFICATION IN ULTRASOUND IMAGING USING COST-SENSITIVE LEARNING AND K-MEANS SMOTE ON THE IMBALANCED BUSI DATASET WITH DEEP FEATURE EXTRACTION

Abstract

Breast cancer is one of the five most common types of cancer that occurs when breast tissue turns into a tumor and mainly affects women. Early diagnosis of the disease is crucial for the patient's lifespan. However, misclassification of malignancy may result in treatment delays and initiate an irreversible process for the patient. This study proposes an approach for classifying ultrasound breast images into malignant, benign, and healthy categories, with a particular emphasis on minimizing false-negative outcomes. The BUSI dataset, characterized by imbalanced class distributions, was used for the breast cancer detection. The dataset was augmented to enhance feature representations using contrast-limited adaptive histogram equalization (CLAHE) to address the class imbalance issue, creating the BUSICL dataset. Features extracted from both datasets with the VGG16 and ResNet50 models were then classified using a support vector machine (SVM). Following the results analysis, the SVM algorithm's cost matrix values were adjusted according to the inverse proportions of class distributions applying a cost-sensitive approach. In addition, the robustness of the proposed methodology is compared with the K-Means SMOTE algorithm. The proposed method achieved an overall accuracy of 99.36%, surpassing the performance of previous comprehensive classification studies using the BUSI dataset.