SEMI-SUPERVISED GAN USING ATTENTION MECHANISM FOR BREAST CANCER CLASSIFICATION

Deep learning approaches have recently contributed to a significant advancement in the field of computer vision. Furthermore, a large-scale annotated dataset is critical for a robust training process. However, obtaining such datasets in the medical field is extremely difficult. In this paper, we use semi-supervised generative adversarial network (SGAN) to describe histological images of breast cancer. SGAN compensates for the lack of training data by making extremely fake images. We add an attention module to SGAN that improves its accuracy by collecting global dependencies and extracting key features. The proposed model is also stabilize using spectral normalization. On the invasive duct carcinoma (IDC) dataset, we demonstrate that our model achieves ۹۴.۳۳%classification accuracy, which is better than the initial semi-supervised SGAN, despite the fact that just ۲% of the data are labeled

کلیدواژه ها:

breast cancer ، image classification ، semi-supervised learning ، generative adversarial networks ، attention mechanism ، invasive ductal carcinoma