_images/autoqc_logo.png _images/autoqc_logo.png

AutoQC#

An automated quality control pipeline for chest radiographs.

Version: 0.7.1

Designed for use prior to training a machine learning model or performing inference, AutoQC aims to:

  • Expedite quality control on chest radiograph datasets, giving developers rapid insights into their data;

  • Improve downstream model generalisation; and

  • Reduce the influence of confounding factors and contribute towards ethical AI.

In pursuit of these goals, AutoQC will:

  1. Perform initial preprocessing;

  2. Detect images that may not be suitable for training or inference; and

  3. Detect and label images for possible confounding factors, e.g. radiographic projection or the presence of a pacemaker.

A pre-print of our paper is available on request:

Automated Quality Control of Chest Radiographs: Tools to Combat Shortcut Learning in COVID-19 Deep Learning Models. (submitted)

Selby IA., González Solares E., Breger A., et al. on behalf of the AIX-COVNET Collaboration.