Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement
In response to the COVID-19 global pandemic, recent research has proposed creating deep learning based models that use chest radiographs (CXRs) in a variety of clinical tasks to help manage the crisis. However, the size of existing datasets of CXRs from COVID-19+ patients are relatively small, and r...
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Public Library of Science (PLoS)
2022-01-01
|
| Series: | PLoS ONE |
| Online Access: | https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9536609/?tool=EBI |
| _version_ | 1817986453351694336 |
|---|---|
| author | Anusua Trivedi Caleb Robinson Marian Blazes Anthony Ortiz Jocelyn Desbiens Sunil Gupta Rahul Dodhia Pavan K. Bhatraju W. Conrad Liles Jayashree Kalpathy-Cramer Aaron Y. Lee Juan M. Lavista Ferres |
| author_facet | Anusua Trivedi Caleb Robinson Marian Blazes Anthony Ortiz Jocelyn Desbiens Sunil Gupta Rahul Dodhia Pavan K. Bhatraju W. Conrad Liles Jayashree Kalpathy-Cramer Aaron Y. Lee Juan M. Lavista Ferres |
| author_sort | Anusua Trivedi |
| collection | DOAJ |
| description | In response to the COVID-19 global pandemic, recent research has proposed creating deep learning based models that use chest radiographs (CXRs) in a variety of clinical tasks to help manage the crisis. However, the size of existing datasets of CXRs from COVID-19+ patients are relatively small, and researchers often pool CXR data from multiple sources, for example, using different x-ray machines in various patient populations under different clinical scenarios. Deep learning models trained on such datasets have been shown to overfit to erroneous features instead of learning pulmonary characteristics in a phenomenon known as shortcut learning. We propose adding feature disentanglement to the training process. This technique forces the models to identify pulmonary features from the images and penalizes them for learning features that can discriminate between the original datasets that the images come from. We find that models trained in this way indeed have better generalization performance on unseen data; in the best case we found that it improved AUC by 0.13 on held out data. We further find that this outperforms masking out non-lung parts of the CXRs and performing histogram equalization, both of which are recently proposed methods for removing biases in CXR datasets. |
| first_indexed | 2024-04-14T00:09:20Z |
| format | Article |
| id | doaj.art-2634f591adf04673a924fc696c4f8847 |
| institution | Directory Open Access Journal |
| issn | 1932-6203 |
| language | English |
| last_indexed | 2024-04-14T00:09:20Z |
| publishDate | 2022-01-01 |
| publisher | Public Library of Science (PLoS) |
| record_format | Article |
| series | PLoS ONE |
| spelling | doaj.art-2634f591adf04673a924fc696c4f88472022-12-22T02:23:24ZengPublic Library of Science (PLoS)PLoS ONE1932-62032022-01-011710Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglementAnusua TrivediCaleb RobinsonMarian BlazesAnthony OrtizJocelyn DesbiensSunil GuptaRahul DodhiaPavan K. BhatrajuW. Conrad LilesJayashree Kalpathy-CramerAaron Y. LeeJuan M. Lavista FerresIn response to the COVID-19 global pandemic, recent research has proposed creating deep learning based models that use chest radiographs (CXRs) in a variety of clinical tasks to help manage the crisis. However, the size of existing datasets of CXRs from COVID-19+ patients are relatively small, and researchers often pool CXR data from multiple sources, for example, using different x-ray machines in various patient populations under different clinical scenarios. Deep learning models trained on such datasets have been shown to overfit to erroneous features instead of learning pulmonary characteristics in a phenomenon known as shortcut learning. We propose adding feature disentanglement to the training process. This technique forces the models to identify pulmonary features from the images and penalizes them for learning features that can discriminate between the original datasets that the images come from. We find that models trained in this way indeed have better generalization performance on unseen data; in the best case we found that it improved AUC by 0.13 on held out data. We further find that this outperforms masking out non-lung parts of the CXRs and performing histogram equalization, both of which are recently proposed methods for removing biases in CXR datasets.https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9536609/?tool=EBI |
| spellingShingle | Anusua Trivedi Caleb Robinson Marian Blazes Anthony Ortiz Jocelyn Desbiens Sunil Gupta Rahul Dodhia Pavan K. Bhatraju W. Conrad Liles Jayashree Kalpathy-Cramer Aaron Y. Lee Juan M. Lavista Ferres Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement PLoS ONE |
| title | Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement |
| title_full | Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement |
| title_fullStr | Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement |
| title_full_unstemmed | Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement |
| title_short | Deep learning models for COVID-19 chest x-ray classification: Preventing shortcut learning using feature disentanglement |
| title_sort | deep learning models for covid 19 chest x ray classification preventing shortcut learning using feature disentanglement |
| url | https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9536609/?tool=EBI |
| work_keys_str_mv | AT anusuatrivedi deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT calebrobinson deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT marianblazes deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT anthonyortiz deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT jocelyndesbiens deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT sunilgupta deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT rahuldodhia deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT pavankbhatraju deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT wconradliles deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT jayashreekalpathycramer deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT aaronylee deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement AT juanmlavistaferres deeplearningmodelsforcovid19chestxrayclassificationpreventingshortcutlearningusingfeaturedisentanglement |