PairDX: A balanced dataset of image–caption pairs spanning six medical document classes

PairDx is a balanced multimodal dataset designed for medical image-text classification tasks. Curated from the well-known MultiCaRe dataset, PairDx comprises 22,665 image-caption pairs spanning six distinct medical categories. This dataset is ideal for researchers and practitioners aiming to develop, test, and benchmark machine learning models that integrate both visual and textual information in the medical domain. The dataset is structured to provide a balanced representation across the six categories. Each image is paired with a corresponding caption that provides a textual description of the medical content, making it suitable for tasks such as image classification, text generation, and cross-modal retrieval. Features: Number of Pairs: 22,665 image-caption pairs. Medical Categories: Six distinct categories relevant to medical image analysis. Image Format: WEBP (high-quality, annotated medical images). Caption Format: Textual descriptions in English. Use Cases: Medical image classification. Cross-modal retrieval (image-to-text, text-to-image). Development of multimodal AI systems in healthcare. Benchmarking image-captioning models in the medical field. License: The dataset is freely available under the MIT License, encouraging open research and innovation in medical AI applications. The dataset can be accessed at [Zenodo DOI link]. Cite as:If you use this dataset in your research, please cite the following paper:Hosam Lafi, H. A. (2025). PairDX: A balanced dataset of image–caption pairs spanning six medical document classes (1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.17452236