Deep Learning Applied to Chest X-Rays: Exploiting and Preventing Shortcuts

Sarah Jabbour, David Fouhey, Ella Kazerooni, Michael W. Sjoding, Jenna Wiens
Proceedings of the 5th Machine Learning for Healthcare Conference, PMLR 126:750-782, 2020.

Abstract

While deep learning has shown promise in improving the automated diagnosis of disease based on chest X-rays, deep networks may exhibit undesirable behavior related to short-cuts. This paper studies the case of spurious class skew in which patients with a particular attribute are spuriously more likely to have the outcome of interest. For instance, clinical protocols might lead to a dataset in which patients with pacemakers are disproportionately likely to have congestive heart failure. This skew can lead to models that take shortcuts by heavily relying on the biased attribute. We explore this problem across a number of attributes in the context of diagnosing the cause of acute hypoxemic respiratory failure. Applied to chest X-rays, we show that i) deep nets can accurately identify many patient attributes including sex (AUROC = 0.96) and age (AUROC 0.90), ii) they tend to exploit correlations between such attributes and the outcome label when learning to predict a diagnosis, leading to poor performance when such correlations do not hold in the test population (e.g., everyone in the test set is male), and iii) a simple transfer learning approach is surprisingly effective at preventing the shortcut and promoting good generalization performance. On the task of diagnosing congestive heart failure based on a set of chest X-rays skewed towards older patients (age $\geq$ 63), the proposed approach improves generalization over standard training from 0.66 (95% CI: 0.54-0.77) to 0.84 (95% CI: 0.73-0.92) AUROC. While simple, the proposed approach has the potential to improve the performance of models across populations by encouraging reliance on clinically relevant manifestations of disease, i.e., those that a clinician would use to make a diagnosis.

Cite this Paper

BibTeX
@InProceedings{pmlr-v126-jabbour20a, title = {Deep Learning Applied to Chest X-Rays: Exploiting and Preventing Shortcuts}, author = {Jabbour, Sarah and Fouhey, David and Kazerooni, Ella and Sjoding, Michael W. and Wiens, Jenna}, booktitle = {Proceedings of the 5th Machine Learning for Healthcare Conference}, pages = {750--782}, year = {2020}, editor = {Doshi-Velez, Finale and Fackler, Jim and Jung, Ken and Kale, David and Ranganath, Rajesh and Wallace, Byron and Wiens, Jenna}, volume = {126}, series = {Proceedings of Machine Learning Research}, month = {07--08 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v126/jabbour20a/jabbour20a.pdf}, url = {https://proceedings.mlr.press/v126/jabbour20a.html}, abstract = {While deep learning has shown promise in improving the automated diagnosis of disease based on chest X-rays, deep networks may exhibit undesirable behavior related to short-cuts. This paper studies the case of spurious class skew in which patients with a particular attribute are spuriously more likely to have the outcome of interest. For instance, clinical protocols might lead to a dataset in which patients with pacemakers are disproportionately likely to have congestive heart failure. This skew can lead to models that take shortcuts by heavily relying on the biased attribute. We explore this problem across a number of attributes in the context of diagnosing the cause of acute hypoxemic respiratory failure. Applied to chest X-rays, we show that i) deep nets can accurately identify many patient attributes including sex (AUROC = 0.96) and age (AUROC 0.90), ii) they tend to exploit correlations between such attributes and the outcome label when learning to predict a diagnosis, leading to poor performance when such correlations do not hold in the test population (e.g., everyone in the test set is male), and iii) a simple transfer learning approach is surprisingly effective at preventing the shortcut and promoting good generalization performance. On the task of diagnosing congestive heart failure based on a set of chest X-rays skewed towards older patients (age $\geq$ 63), the proposed approach improves generalization over standard training from 0.66 (95% CI: 0.54-0.77) to 0.84 (95% CI: 0.73-0.92) AUROC. While simple, the proposed approach has the potential to improve the performance of models across populations by encouraging reliance on clinically relevant manifestations of disease, i.e., those that a clinician would use to make a diagnosis.} }
Endnote
%0 Conference Paper %T Deep Learning Applied to Chest X-Rays: Exploiting and Preventing Shortcuts %A Sarah Jabbour %A David Fouhey %A Ella Kazerooni %A Michael W. Sjoding %A Jenna Wiens %B Proceedings of the 5th Machine Learning for Healthcare Conference %C Proceedings of Machine Learning Research %D 2020 %E Finale Doshi-Velez %E Jim Fackler %E Ken Jung %E David Kale %E Rajesh Ranganath %E Byron Wallace %E Jenna Wiens %F pmlr-v126-jabbour20a %I PMLR %P 750--782 %U https://proceedings.mlr.press/v126/jabbour20a.html %V 126 %X While deep learning has shown promise in improving the automated diagnosis of disease based on chest X-rays, deep networks may exhibit undesirable behavior related to short-cuts. This paper studies the case of spurious class skew in which patients with a particular attribute are spuriously more likely to have the outcome of interest. For instance, clinical protocols might lead to a dataset in which patients with pacemakers are disproportionately likely to have congestive heart failure. This skew can lead to models that take shortcuts by heavily relying on the biased attribute. We explore this problem across a number of attributes in the context of diagnosing the cause of acute hypoxemic respiratory failure. Applied to chest X-rays, we show that i) deep nets can accurately identify many patient attributes including sex (AUROC = 0.96) and age (AUROC 0.90), ii) they tend to exploit correlations between such attributes and the outcome label when learning to predict a diagnosis, leading to poor performance when such correlations do not hold in the test population (e.g., everyone in the test set is male), and iii) a simple transfer learning approach is surprisingly effective at preventing the shortcut and promoting good generalization performance. On the task of diagnosing congestive heart failure based on a set of chest X-rays skewed towards older patients (age $\geq$ 63), the proposed approach improves generalization over standard training from 0.66 (95% CI: 0.54-0.77) to 0.84 (95% CI: 0.73-0.92) AUROC. While simple, the proposed approach has the potential to improve the performance of models across populations by encouraging reliance on clinically relevant manifestations of disease, i.e., those that a clinician would use to make a diagnosis.
APA
Jabbour, S., Fouhey, D., Kazerooni, E., Sjoding, M.W. & Wiens, J.. (2020). Deep Learning Applied to Chest X-Rays: Exploiting and Preventing Shortcuts. Proceedings of the 5th Machine Learning for Healthcare Conference, in Proceedings of Machine Learning Research 126:750-782 Available from https://proceedings.mlr.press/v126/jabbour20a.html.

Related Material