PMC-CLIP: Contrastive Language-Image Pre-training using Biomedical Documents
This addresses data scarcity in biomedical AI, enabling better multimodal models for medical applications, though it is incremental as it adapts an existing CLIP-style approach to a new domain.
The authors tackled the lack of large-scale biomedical datasets for foundation models by creating PMC-OA, a dataset with 1.6M image-caption pairs, and trained PMC-CLIP, which achieved state-of-the-art results, including +8.1% R@10 on image-text retrieval and +3.9% accuracy on image classification.
Foundation models trained on large-scale dataset gain a recent surge in CV and NLP. In contrast, development in biomedical domain lags far behind due to data scarcity. To address this issue, we build and release PMC-OA, a biomedical dataset with 1.6M image-caption pairs collected from PubMedCentral's OpenAccess subset, which is 8 times larger than before. PMC-OA covers diverse modalities or diseases, with majority of the image-caption samples aligned at finer-grained level, i.e., subfigure and subcaption. While pretraining a CLIP-style model on PMC-OA, our model named PMC-CLIP achieves state-of-the-art results on various downstream tasks, including image-text retrieval on ROCO, MedMNIST image classification, Medical VQA, i.e. +8.1% R@10 on image-text retrieval, +3.9% accuracy on image classification.