◐ModelHealthcareFree
TorchXRayVision
Pretrained chest X-ray classification, segmentation, and autoencoder models with unified loaders across major public CXR datasets.
TorchXRayVision
A library of pretrained chest-radiograph models and standardized dataset loaders. It unifies a dozen public CXR datasets (NIH, CheXpert, MIMIC-CXR, PadChest and others) behind one interface with a common label space, and ships DenseNet/ResNet classifiers plus autoencoders and a lung segmentation model that work across sources. It targets a recurring pain in medical imaging research: every dataset labels pathologies differently, so reproducible cross-dataset baselines are hard to build.
Key features
- Pretrained multi-label classifiers for 18 common chest pathologies
- Unified loaders that harmonize NIH, CheXpert, MIMIC-CXR, PadChest and more into one label space
- Pretrained lung segmentation and image autoencoder models
- Baseline utilities for cross-dataset generalization and calibration studies
- Pure PyTorch, so models drop straight into existing training and inference code
Curated mirror of the open-source TorchXRayVision (Apache-2.0). Get it from the source.