MedCLIP: Contrastive Learning from Unpaired Medical Images and Text
This addresses the challenge of multimodal learning in medical AI where data is scarce and noisy, offering an incremental improvement over existing methods.
The paper tackles the problem of limited and noisy medical image-text datasets for contrastive learning by decoupling images and texts to scale training data and using a semantic matching loss to reduce false negatives, achieving state-of-the-art results in zero-shot prediction, classification, and retrieval with only 20K pre-training data.
Existing vision-text contrastive learning like CLIP aims to match the paired image and caption embeddings while pushing others apart, which improves representation transferability and supports zero-shot prediction. However, medical image-text datasets are orders of magnitude below the general images and captions from the internet. Moreover, previous methods encounter many false negatives, i.e., images and reports from separate patients probably carry the same semantics but are wrongly treated as negatives. In this paper, we decouple images and texts for multimodal contrastive learning thus scaling the usable training data in a combinatorial magnitude with low cost. We also propose to replace the InfoNCE loss with semantic matching loss based on medical knowledge to eliminate false negatives in contrastive learning. We prove that MedCLIP is a simple yet effective framework: it outperforms state-of-the-art methods on zero-shot prediction, supervised classification, and image-text retrieval. Surprisingly, we observe that with only 20K pre-training data, MedCLIP wins over the state-of-the-art method (using around 200K data). Our code is available at https://github.com/RyanWangZf/MedCLIP.