Deep Learning based Segmentation and Classification of Breast Ultrasound Images

Breast cancer remains a major public health concern, requiring better diagnostic technologies for early identification and treatment. In this research, we leverage the Breast Ultrasound Image dataset (BUSI) to construct a comprehensive approach to breast cancer detection employing deep learning methodologies. We employ the U-Net architecture for segmentation, training the model on ultrasound images along with their corresponding masked images to automatically delineate tumor regions. Additionally, we explore classification using state-of-the-art models, including EfficientNetV2-S, ResNeXt-101 (64*4d), and MaxVit, initially utilizing solely ultrasound images. Subsequently, we enhance classification accuracy by overlaying the segmented masked images onto the ultrasound images, augmenting the feature representation for classification. Our results demonstrate the effectiveness of integrating segmentation and classification approaches, yielding improved accuracy in breast cancer detection compared to standalone classification methods. In the diagnosis of breast cancer, this combined strategy has the potential to enhance clinical decision-making and patient outcomes.