Addressing Chest Radiograph Projection Bias in Deep Classification Models

Sofia Cardoso Pereira, Joana Rocha, Alex Gaudio, Asim Smailagic, Aur’elio Campilho, Ana Maria Mendonça
Medical Imaging with Deep Learning, PMLR 227:1199-1210, 2024.

Abstract

Deep learning-based models are widely used for disease classification in chest radiographs. This exam can be performed in one of two projections (posteroanterior or anteroposterior), depending on the direction that the X-ray beam travels through the body. Since projection visibly affects the way anatomical structures appear in the scans, it may introduce bias in classifiers, especially when spurious correlations between a given disease and a projection occur. This paper examines the influence of chest radiograph projection on the performance of deep learning-based classification models and proposes an approach to mitigate projection-induced bias. Results show that a DenseNet-121 model is better at classifying images from the most representative projection in the data set, suggesting that projection is taken into account by the classifier. Moreover, this model can classify chest X-ray projection better than any of the fourteen radiological findings considered, without being explicitly trained for that task, putting it at high risk for projection bias. We propose a label-conditional gradient reversal framework to make the model insensitive to projection, by forcing the extracted features to be simultaneously good for disease classification and bad for projection classification, resulting in a framework with reduced projection-induced bias.

Cite this Paper

BibTeX
@InProceedings{pmlr-v227-pereira24a, title = {Addressing Chest Radiograph Projection Bias in Deep Classification Models}, author = {Pereira, Sofia Cardoso and Rocha, Joana and Gaudio, Alex and Smailagic, Asim and Campilho, Aur'elio and Mendon\c{c}a, Ana Maria}, booktitle = {Medical Imaging with Deep Learning}, pages = {1199--1210}, year = {2024}, editor = {Oguz, Ipek and Noble, Jack and Li, Xiaoxiao and Styner, Martin and Baumgartner, Christian and Rusu, Mirabela and Heinmann, Tobias and Kontos, Despina and Landman, Bennett and Dawant, Benoit}, volume = {227}, series = {Proceedings of Machine Learning Research}, month = {10--12 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v227/pereira24a/pereira24a.pdf}, url = {https://proceedings.mlr.press/v227/pereira24a.html}, abstract = {Deep learning-based models are widely used for disease classification in chest radiographs. This exam can be performed in one of two projections (posteroanterior or anteroposterior), depending on the direction that the X-ray beam travels through the body. Since projection visibly affects the way anatomical structures appear in the scans, it may introduce bias in classifiers, especially when spurious correlations between a given disease and a projection occur. This paper examines the influence of chest radiograph projection on the performance of deep learning-based classification models and proposes an approach to mitigate projection-induced bias. Results show that a DenseNet-121 model is better at classifying images from the most representative projection in the data set, suggesting that projection is taken into account by the classifier. Moreover, this model can classify chest X-ray projection better than any of the fourteen radiological findings considered, without being explicitly trained for that task, putting it at high risk for projection bias. We propose a label-conditional gradient reversal framework to make the model insensitive to projection, by forcing the extracted features to be simultaneously good for disease classification and bad for projection classification, resulting in a framework with reduced projection-induced bias.} }
Endnote
%0 Conference Paper %T Addressing Chest Radiograph Projection Bias in Deep Classification Models %A Sofia Cardoso Pereira %A Joana Rocha %A Alex Gaudio %A Asim Smailagic %A Aur’elio Campilho %A Ana Maria Mendonça %B Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2024 %E Ipek Oguz %E Jack Noble %E Xiaoxiao Li %E Martin Styner %E Christian Baumgartner %E Mirabela Rusu %E Tobias Heinmann %E Despina Kontos %E Bennett Landman %E Benoit Dawant %F pmlr-v227-pereira24a %I PMLR %P 1199--1210 %U https://proceedings.mlr.press/v227/pereira24a.html %V 227 %X Deep learning-based models are widely used for disease classification in chest radiographs. This exam can be performed in one of two projections (posteroanterior or anteroposterior), depending on the direction that the X-ray beam travels through the body. Since projection visibly affects the way anatomical structures appear in the scans, it may introduce bias in classifiers, especially when spurious correlations between a given disease and a projection occur. This paper examines the influence of chest radiograph projection on the performance of deep learning-based classification models and proposes an approach to mitigate projection-induced bias. Results show that a DenseNet-121 model is better at classifying images from the most representative projection in the data set, suggesting that projection is taken into account by the classifier. Moreover, this model can classify chest X-ray projection better than any of the fourteen radiological findings considered, without being explicitly trained for that task, putting it at high risk for projection bias. We propose a label-conditional gradient reversal framework to make the model insensitive to projection, by forcing the extracted features to be simultaneously good for disease classification and bad for projection classification, resulting in a framework with reduced projection-induced bias.
APA
Pereira, S.C., Rocha, J., Gaudio, A., Smailagic, A., Campilho, A. & Mendonça, A.M.. (2024). Addressing Chest Radiograph Projection Bias in Deep Classification Models. Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 227:1199-1210 Available from https://proceedings.mlr.press/v227/pereira24a.html.

Related Material