Explainable Image Quality Analysis of Chest X-Rays

Caner Ozer, Ilkay Oksuz
Proceedings of the Fourth Conference on Medical Imaging with Deep Learning, PMLR 143:567-580, 2021.

Abstract

Medical image quality assessment is an important aspect of image acquisition where poor-quality images may lead to misdiagnosis. In addition, manual labelling of image quality after the acquisition is often tedious and can lead to some misleading results. Despite much research on the automated analysis of image quality for tackling this problem, relatively little work has been done for the explanation of the methodologies. In this work, we propose an explainable image quality assessment system and validate our idea on foreign objects in a Chest X-Ray (Object-CXR) dataset. Our explainable pipeline relies on NormGrad, an algorithm, which can efficiently localize the image quality issues with saliency maps of the classifier. We compare our method with a range of saliency detection methods and illustrate the superior performance of NormGrad by obtaining a Pointing Game accuracy of 0.862 on the test dataset of the Object-CXR dataset. We also verify our findings through a qualitative analysis by visualizing attention maps for foreign objects on X-Ray images.

Cite this Paper

BibTeX
@InProceedings{pmlr-v143-ozer21a, title = {Explainable Image Quality Analysis of Chest X-Rays}, author = {Ozer, Caner and Oksuz, Ilkay}, booktitle = {Proceedings of the Fourth Conference on Medical Imaging with Deep Learning}, pages = {567--580}, year = {2021}, editor = {Heinrich, Mattias and Dou, Qi and de Bruijne, Marleen and Lellmann, Jan and Schläfer, Alexander and Ernst, Floris}, volume = {143}, series = {Proceedings of Machine Learning Research}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v143/ozer21a/ozer21a.pdf}, url = {https://proceedings.mlr.press/v143/ozer21a.html}, abstract = {Medical image quality assessment is an important aspect of image acquisition where poor-quality images may lead to misdiagnosis. In addition, manual labelling of image quality after the acquisition is often tedious and can lead to some misleading results. Despite much research on the automated analysis of image quality for tackling this problem, relatively little work has been done for the explanation of the methodologies. In this work, we propose an explainable image quality assessment system and validate our idea on foreign objects in a Chest X-Ray (Object-CXR) dataset. Our explainable pipeline relies on NormGrad, an algorithm, which can efficiently localize the image quality issues with saliency maps of the classifier. We compare our method with a range of saliency detection methods and illustrate the superior performance of NormGrad by obtaining a Pointing Game accuracy of 0.862 on the test dataset of the Object-CXR dataset. We also verify our findings through a qualitative analysis by visualizing attention maps for foreign objects on X-Ray images.} }
Endnote
%0 Conference Paper %T Explainable Image Quality Analysis of Chest X-Rays %A Caner Ozer %A Ilkay Oksuz %B Proceedings of the Fourth Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2021 %E Mattias Heinrich %E Qi Dou %E Marleen de Bruijne %E Jan Lellmann %E Alexander Schläfer %E Floris Ernst %F pmlr-v143-ozer21a %I PMLR %P 567--580 %U https://proceedings.mlr.press/v143/ozer21a.html %V 143 %X Medical image quality assessment is an important aspect of image acquisition where poor-quality images may lead to misdiagnosis. In addition, manual labelling of image quality after the acquisition is often tedious and can lead to some misleading results. Despite much research on the automated analysis of image quality for tackling this problem, relatively little work has been done for the explanation of the methodologies. In this work, we propose an explainable image quality assessment system and validate our idea on foreign objects in a Chest X-Ray (Object-CXR) dataset. Our explainable pipeline relies on NormGrad, an algorithm, which can efficiently localize the image quality issues with saliency maps of the classifier. We compare our method with a range of saliency detection methods and illustrate the superior performance of NormGrad by obtaining a Pointing Game accuracy of 0.862 on the test dataset of the Object-CXR dataset. We also verify our findings through a qualitative analysis by visualizing attention maps for foreign objects on X-Ray images.
APA
Ozer, C. & Oksuz, I.. (2021). Explainable Image Quality Analysis of Chest X-Rays. Proceedings of the Fourth Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 143:567-580 Available from https://proceedings.mlr.press/v143/ozer21a.html.

Related Material