Manifestation-Guided Multimodal Pretraining for Improved Mammography Classification
Leveraging manifestations as semantic proxies, the ManiNeg framework enhances hard negative sampling in contrastive learning, leading to more informative representations for improved mammography classification.