MedMNIST: 18 Standardized 2D and 3D Biomedical Image Classification Subsets

MedMNIST consists of 18 standardized subsets for 2D and 3D biomedical image classification, maintained by the MedMNIST organization and updated as of January 2025. It provides a large-scale benchmark for medical image analysis across multiple modalities including 2D and 3D data.

Use Cases

• Training 3D convolutional neural networks on the 3D subsets for volumetric classification
• Benchmarking AutoML frameworks across 18 distinct medical imaging modalities
• Evaluating few-shot learning performance using the standardized image formats

Strengths

• 18 standardized subsets for benchmarking
• Includes both 2D and 3D image modalities
• Apache-2.0 licensed for open research

Limitations

• Low-resolution MNIST-style formatting may omit fine-grained clinical details
• Potential class imbalance within specific subsets

Provenance

Source

MedMNIST Organization

Collection Method

Standardization of existing biomedical imaging datasets

Freshness

Updated January 2025