BreastUS: Vision Transformer for Breast Cancer Classification Using Breast Ultrasound Images
A transformer model, namely BreastUS Transformer, for Breast ultrasound Image Classification (BUIC), which incorporates self-attention and enables the automatic classification of breast ultrasound images into normal, benign, and malignant cases is proposed.
Abstract
Breast cancer is the most common and fatal cancer type in women, posing a serious threat to women’s health. Ultrasounds play an essential role in the testing and diagnosis of breast cancer. Currently, it is the golden standard for complimentary screening for dense breasts, which has the potential to develop tumours. This paper proposes a transformer model, namely BreastUS Transformer, for Breast ultrasound Image Classification (BUIC). BreastUS incorporates self-attention and enables the automatic classification of breast ultrasound images into normal, benign, and malignant cases. The performance of BrestUS is evaluated using standard performance metrics like accuracy, recall, precision, and the F1-score. The quantitative results are compared with five state-of-the-art convolutional neural networks (CNN), and at least a 4% performance boost is achieved compared to the best-performing CNN.