Leveraging manifestations as semantic proxies, the ManiNeg framework enhances hard negative sampling in contrastive learning, leading to more informative representations for improved mammography classification.
