Do comprehensive deep learning algorithms suffer from hidden stratification? A retrospective study on pneumothorax detection in chest radiography.

Journal: BMJ open
Published Date: Dec 7, 2021

Abstract

OBJECTIVES: To evaluate the ability of a commercially available comprehensive chest radiography deep convolutional neural network (DCNN) to detect simple and tension pneumothorax, as stratified by the following subgroups: the presence of an intercostal drain; rib, clavicular, scapular or humeral fractures or rib resections; subcutaneous emphysema and erect versus non-erect positioning. The hypothesis was that performance would not differ significantly in each of these subgroups when compared with the overall test dataset.

Authors

  • Jarrel Seah
    Department of Neuroscience, Monash University, Melbourne, Australia; Radiology and Nuclear Medicine, Alfred Health, Melbourne, Australia.
  • Cyril Tang
    Annalise-AI, Sydney, New South Wales, Australia.
  • Quinlan D Buchlak
    School of Medicine, The University of Notre Dame, Sydney, NSW, Australia. quinlan.buchlak1@my.nd.edu.au.
  • Michael Robert Milne
    annalise.ai, Sydney, New South Wales, Australia.
  • Xavier Holt
    annalise.ai, Sydney, New South Wales, Australia.
  • Hassan Ahmad
    IBM Almaden Research Center, San Jose, CA.
  • John Lambert
    annalise.ai, Sydney, New South Wales, Australia.
  • Nazanin Esmaili
    School of Medicine, The University of Notre Dame, Sydney, NSW, Australia.
  • Luke Oakden-Rayner
    Department of Medical Imaging Research, Royal Adelaide Hospital, Adelaide, Australia.
  • Peter Brotchie
    St Vincent's Hospital, Melbourne, Victoria, Australia.
  • Catherine M Jones
    I-MED Radiology Network, Brisbane, Queensland, Australia.

Keywords

External Resources

Popular Topics
Recent Journals