BREAST CANCER CLASSIFICATION SYSTEM USING ULTRASOUND IMAGES

Abstract

Breast cancer is one of the most common and life-threatening cancers in women worldwide. Early detection and accurate diagnosis are critical in reducing mortality rates. Ultrasound imaging is widely used in clinical settings for its accessibility and effectiveness in detecting abnormalities in breast tissue. However, manual interpretation of ultrasound images is time-consuming and prone to errors. In this study, a deep learning-based Breast Cancer Detection and Classification System is developed to assist in identifying and categorizing breast tumors into benign or malignant types. Two approaches are explored in this research. The first approach employs a single model to detect tumor locations and classify them simultaneously. The second approach utilizes two specialized models: one for tumor detection and the other for classification. A publicly available dataset of 762 ultrasound images, categorized into benign, malignant, and normal classes, was augmented to approximately 1,800 images to enhance model performance. The dataset was divided into subsets to train and evaluate the models. Evaluation metrics such as precision, recall, F1-score, and mean Average Precision (mAP) were used to assess model performance. Results indicate that the single-model approach performed well for benign tumors but struggled with malignant classification. The two-model approach showed better performance in balancing precision and recall for both tumor types. This study highlights the potential of deep learning in enhancing breast cancer diagnosis and provides a foundation for further development of computer-aided diagnostic systems in medical imaging. Future work will focus on optimizing the models and exploring additional datasets to improve classification accuracy and generalization.