MedIMeta: A comprehensive and easy-to-use multi-domain multi-task medical imaging meta-dataset

We introduce the Medical Imaging Meta-Dataset (MedIMeta), a novel multi-domain, multi-task meta-dataset designed to facilitate the development and standardised evaluation of ML models and cross-domain few-shot learning algorithms for medical image classification. MedIMeta contains 19 medical imaging datasets spanning 10 different domains and encompassing 54 distinct medical tasks, offering opportunities for both single-task and multi-task training. All tasks are standardised to the same format and readily usable in PyTorch or other ML frameworks. All datasets have been previously published with an open license that allows redistribution or we obtained an explicit permission to do so.

Each dataset within the MedIMeta dataset is standardized to a size of 224 × 224 pixels which matches image size commonly used in pre-trained models. Furthermore, the dataset comes with pre-made splits to ensure ease of use and standardized benchmarking. We release a user-friendly Python package to directly load images for use in PyTorch.

Links

• Project website: https://www.woerner.eu/projects/medimeta/
• Data loading code (medimeta Python package): https://github.com/StefanoWoerner/medimeta-pytorch
• Data creation code: https://github.com/StefanoWoerner/medimeta-dataset-scripts

Dataset Overview