Mitigating annotation shift in cancer classification using single-image generative models | Publicación