ROCOv2 Radiology Multimodal Medical Image Dataset

ROCOv2 (Radiology Object in COntext Version 2) is an innovative multimodal medical image dataset that combines radiology images with related medical concepts and descriptions. This dataset extracts radiology images and related medical concepts and descriptions from the PMC Open Access subset. It is an updated version of the ROCO dataset, adding 35,705 new images and improving concept extraction and filtering.

The dataset contains 79,789 radiology images covering a variety of clinical modalities, anatomical regions, and orientations (for X-rays), each with a corresponding medical concept description.

The uses of the ROCOv2 dataset include: training image annotation models based on image-caption pairs, multi-label image classification using UMLS concepts, pre-training of medical domain models, evaluation of deep learning models in multi-task learning, image retrieval and caption generation tasks. In addition, the application value of the ROCOv2 dataset in the medical field lies in the fact that it contains a large amount of biomedical knowledge, which can be used to train and evaluate models of different modalities, such as image caption generation, as well as to build and train efficient image retrieval systems specifically for the medical field.