MonkeyPox Skin Image Dataset for Computer Vision (MPox-Vision)
Description
Mpox-Vision is a quality-verified dataset of monkeypox skin images curated to support the development of reliable AI-based diagnostic systems. All images have been reviewed by two medical experts to ensure proper focus, exposure, and clinical relevance. The dataset excludes irrelevant, mislabeled, or noisy images, making it clean and suitable for training robust and generalizable models. Mpox-Vision offers a dependable resource for advancing skin lesion classification and AI-assisted diagnosis, especially in low-resource settings. Please cite the following paper if you use this dataset: Hossain, M. S., Ahmed, M., & Rahman, M. S. (2025). From survey to solution: A deep learning framework for reliable monkeypox diagnosis using skin images. Array, 28, 100554. https://doi.org/10.1016/j.array.2025.100554
Files
Steps to reproduce
In this study, we compiled a new, more realistic dataset following a 5-step process, shown in Fig. 4. Firstly, we collected monkeypox, chick- enpox and measles lesion images from publicly available datasets and through web searches. Unlike previous datasets, we excluded lesion-free healthy skin images and included acne images to introduce greater inter-class similarity and make the classification task more challenging and clinically meaningful. Then, we used Google Reverse Image Search to verify the authenticity of each image. This helped identify duplicate or mislabeled images by cross-referencing their appearance across multiple online sources. Only images with verified sources were retained. Following that, a comprehensive three-stage screening process was applied to rigorously eliminate noise and ensure the quality of the image dataset. At first, we manually removed the irrelevant images and cropped images to remove unnecessary backgrounds, as shown in Figs. 3 (i,j,k,l) and (f,g,h) respectively. We also manually removed images with inconsistent magnifications. We intended to include images in the MPox-Vision dataset with a resolution similar to that of smartphone-captured images since we wanted to develop a method to detect monkeypox on-site from smartphone-captured images. The remaining images were resized to 224 × 224 pixels while preserving the original aspect ratio. Then, out-of-focus, noisy and underexposed images were detected and eliminated using the automated quality evaluation method. For evaluating the focus, exposure and noise of the images, we utilized the quality evaluation method developed for medical images [76]. Finally, two experts carefully reviewed the remaining images and selected 200 images per class, ensuring correct labeling and eliminating any residual noise or mislabeled images. The resulting MPox-Vision dataset is thus more balanced, diverse, and suited for evaluating deep learning-based monkeypox detection methods.