Error-free Training for MedMNIST: Pre-trained Models for Medical Image Datasets

Pre-trained machine learning models for three MedMNIST medical imaging datasets: NoduleMNIST3D, OrganAMNIST, and PathMNIST. The models were trained to 100% accuracy and are discussed in the arXiv paper 'Error-free Training for MedMNIST Datasets'. The dataset was authored by Bo Deng and last updated on April 20, 2026.

Use Cases

• Benchmarking model performance on MedMNIST datasets based on the provided pre-trained models.
• Reproducing results from the 'Error-free Training for MedMNIST Datasets' arXiv paper.
• Using pre-trained models as a baseline for medical image classification tasks on nodules, organs, or pathology slides.
• Studying the properties of models trained to 100% accuracy on medical imaging data.

Strengths

• Models are trained to 100% accuracy on their respective MedMNIST datasets.
• Covers three distinct medical imaging domains: 3D nodules, organ segmentation, and pathology.
• Dataset size is 2.0 GB, indicating substantial model parameter files.
• Released under a permissive CC-BY-4.0 license.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description is focused on model artifacts, not the underlying image data.

Provenance

Source

figshare, authored by Bo Deng.

Collection Method

Models were generated as part of the research discussed in the arXiv paper 'Error-free Training for MedMNIST Datasets'.

Freshness

Last updated 2026-04-20 01:17:51; freshness should be verified.