Deploying deep learning models on unseen medical imaging using adversarial domain adaptation
The fundamental challenge in machine learning is ensuring that trained models generalize well to unseen data. We developed a general technique for ameliorating the effect of dataset shift using generative adversarial networks (GANs) on a dataset of 149,298 handwritten digits and dataset of 868,549 c...
| 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/PMC9565422/?tool=EBI |
| _version_ | 1828344370164662272 |
|---|---|
| author | Aly A. Valliani Faris F. Gulamali Young Joon Kwon Michael L. Martini Chiatse Wang Douglas Kondziolka Viola J. Chen Weichung Wang Anthony B. Costa Eric K. Oermann |
| author_facet | Aly A. Valliani Faris F. Gulamali Young Joon Kwon Michael L. Martini Chiatse Wang Douglas Kondziolka Viola J. Chen Weichung Wang Anthony B. Costa Eric K. Oermann |
| author_sort | Aly A. Valliani |
| collection | DOAJ |
| description | The fundamental challenge in machine learning is ensuring that trained models generalize well to unseen data. We developed a general technique for ameliorating the effect of dataset shift using generative adversarial networks (GANs) on a dataset of 149,298 handwritten digits and dataset of 868,549 chest radiographs obtained from four academic medical centers. Efficacy was assessed by comparing area under the curve (AUC) pre- and post-adaptation. On the digit recognition task, the baseline CNN achieved an average internal test AUC of 99.87% (95% CI, 99.87-99.87%), which decreased to an average external test AUC of 91.85% (95% CI, 91.82-91.88%), with an average salvage of 35% from baseline upon adaptation. On the lung pathology classification task, the baseline CNN achieved an average internal test AUC of 78.07% (95% CI, 77.97-78.17%) and an average external test AUC of 71.43% (95% CI, 71.32-71.60%), with a salvage of 25% from baseline upon adaptation. Adversarial domain adaptation leads to improved model performance on radiographic data derived from multiple out-of-sample healthcare populations. This work can be applied to other medical imaging domains to help shape the deployment toolkit of machine learning in medicine. |
| first_indexed | 2024-04-13T23:56:18Z |
| format | Article |
| id | doaj.art-2a81c920dec8430abc753a92bfe4be4a |
| institution | Directory Open Access Journal |
| issn | 1932-6203 |
| language | English |
| last_indexed | 2024-04-13T23:56:18Z |
| publishDate | 2022-01-01 |
| publisher | Public Library of Science (PLoS) |
| record_format | Article |
| series | PLoS ONE |
| spelling | doaj.art-2a81c920dec8430abc753a92bfe4be4a2022-12-22T02:23:52ZengPublic Library of Science (PLoS)PLoS ONE1932-62032022-01-011710Deploying deep learning models on unseen medical imaging using adversarial domain adaptationAly A. VallianiFaris F. GulamaliYoung Joon KwonMichael L. MartiniChiatse WangDouglas KondziolkaViola J. ChenWeichung WangAnthony B. CostaEric K. OermannThe fundamental challenge in machine learning is ensuring that trained models generalize well to unseen data. We developed a general technique for ameliorating the effect of dataset shift using generative adversarial networks (GANs) on a dataset of 149,298 handwritten digits and dataset of 868,549 chest radiographs obtained from four academic medical centers. Efficacy was assessed by comparing area under the curve (AUC) pre- and post-adaptation. On the digit recognition task, the baseline CNN achieved an average internal test AUC of 99.87% (95% CI, 99.87-99.87%), which decreased to an average external test AUC of 91.85% (95% CI, 91.82-91.88%), with an average salvage of 35% from baseline upon adaptation. On the lung pathology classification task, the baseline CNN achieved an average internal test AUC of 78.07% (95% CI, 77.97-78.17%) and an average external test AUC of 71.43% (95% CI, 71.32-71.60%), with a salvage of 25% from baseline upon adaptation. Adversarial domain adaptation leads to improved model performance on radiographic data derived from multiple out-of-sample healthcare populations. This work can be applied to other medical imaging domains to help shape the deployment toolkit of machine learning in medicine.https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9565422/?tool=EBI |
| spellingShingle | Aly A. Valliani Faris F. Gulamali Young Joon Kwon Michael L. Martini Chiatse Wang Douglas Kondziolka Viola J. Chen Weichung Wang Anthony B. Costa Eric K. Oermann Deploying deep learning models on unseen medical imaging using adversarial domain adaptation PLoS ONE |
| title | Deploying deep learning models on unseen medical imaging using adversarial domain adaptation |
| title_full | Deploying deep learning models on unseen medical imaging using adversarial domain adaptation |
| title_fullStr | Deploying deep learning models on unseen medical imaging using adversarial domain adaptation |
| title_full_unstemmed | Deploying deep learning models on unseen medical imaging using adversarial domain adaptation |
| title_short | Deploying deep learning models on unseen medical imaging using adversarial domain adaptation |
| title_sort | deploying deep learning models on unseen medical imaging using adversarial domain adaptation |
| url | https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9565422/?tool=EBI |
| work_keys_str_mv | AT alyavalliani deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT farisfgulamali deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT youngjoonkwon deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT michaellmartini deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT chiatsewang deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT douglaskondziolka deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT violajchen deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT weichungwang deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT anthonybcosta deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation AT erickoermann deployingdeeplearningmodelsonunseenmedicalimagingusingadversarialdomainadaptation |