TorchXRayVision: A library of chest X-ray datasets and models

TL;DR

The paper introduces TorchXRayVision, an open-source framework that provides a unified interface for numerous chest X-ray datasets and deep learning models to enable reproducible baselines and rapid experimentation. It structures models into core classifiers tailored for the library and external baseline classifiers, with automatic image resizing, downloadable pre-trained weights, and convenient feature extraction capabilities. The framework includes dataset utilities (merge, subset, relabel, view filtering), supports pathology and semantic masks, and offers tools to simulate covariate shifts for robustness and generalization studies. This work facilitates cross-dataset evaluation, transfer learning workflows, and systematic analysis of model failures in medical imaging.

Abstract

TorchXRayVision is an open source software library for working with chest X-ray datasets and deep learning models. It provides a common interface and common pre-processing chain for a wide set of publicly available chest X-ray datasets. In addition, a number of classification and representation learning models with different architectures, trained on different data combinations, are available through the library to serve as baselines or feature extractors.