Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence
Purpose: Developing robust artificial intelligence (AI) models for medical image analysis requires large quantities of diverse, well-chosen data that can prove challenging to collect because of privacy concerns, disease rarity, or diagnostic label quality. Collecting image-based datasets for retinop...
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2022-06-01
|
| Series: | Ophthalmology Science |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S266691452200015X |
| _version_ | 1818249285597134848 |
|---|---|
| author | Aaron S. Coyner, PhD Jimmy S. Chen, MD Ken Chang, PhD Praveer Singh, PhD Susan Ostmo, MS R. V. Paul Chan, MD, MSc Michael F. Chiang, MD, MA Jayashree Kalpathy-Cramer, PhD J. Peter Campbell, MD, MPH |
| author_facet | Aaron S. Coyner, PhD Jimmy S. Chen, MD Ken Chang, PhD Praveer Singh, PhD Susan Ostmo, MS R. V. Paul Chan, MD, MSc Michael F. Chiang, MD, MA Jayashree Kalpathy-Cramer, PhD J. Peter Campbell, MD, MPH |
| author_sort | Aaron S. Coyner, PhD |
| collection | DOAJ |
| description | Purpose: Developing robust artificial intelligence (AI) models for medical image analysis requires large quantities of diverse, well-chosen data that can prove challenging to collect because of privacy concerns, disease rarity, or diagnostic label quality. Collecting image-based datasets for retinopathy of prematurity (ROP), a potentially blinding disease, suffers from these challenges. Progressively growing generative adversarial networks (PGANs) may help, because they can synthesize highly realistic images that may increase both the size and diversity of medical datasets. Design: Diagnostic validation study of convolutional neural networks (CNNs) for plus disease detection, a component of severe ROP, using synthetic data. Participants: Five thousand eight hundred forty-two retinal fundus images (RFIs) collected from 963 preterm infants. Methods: Retinal vessel maps (RVMs) were segmented from RFIs. PGANs were trained to synthesize RVMs with normal, pre-plus, or plus disease vasculature. Convolutional neural networks were trained, using real or synthetic RVMs, to detect plus disease from 2 real RVM test datasets. Main Outcome Measures: Features of real and synthetic RVMs were evaluated using uniform manifold approximation and projection (UMAP). Similarities were evaluated at the dataset and feature level using Fréchet inception distance and Euclidean distance, respectively. CNN performance was assessed via area under the receiver operating characteristic curve (AUC); AUCs were compared via bootstrapping and Delong’s test for correlated receiver operating characteristic curves. Confusion matrices were compared using McNemar’s chi-square test and Cohen’s κ value. Results: The CNN trained on synthetic RVMs showed a significantly higher AUC (0.971; P = 0.006 and P = 0.004) and classified plus disease more similarly to a set of 8 international experts (κ = 0.922) than the CNN trained on real RVMs (AUC = 0.934; κ = 0.701). Real and synthetic RVMs overlapped, by plus disease diagnosis, on the UMAP manifold, showing that synthetic images spanned the disease severity spectrum. Fréchet inception distance and Euclidean distances suggested that real and synthetic RVMs were more dissimilar to one another than real RVMs were to one another, further suggesting that synthetic RVMs were distinct from the training data with respect to privacy considerations. Conclusions: Synthetic datasets may be useful for training robust medical AI models. Furthermore, PGANs may be able to synthesize realistic data for use without protected health information concerns. |
| first_indexed | 2024-12-12T15:34:04Z |
| format | Article |
| id | doaj.art-2e71409d77a341a8b6c30ab2a5e34f77 |
| institution | Directory Open Access Journal |
| issn | 2666-9145 |
| language | English |
| last_indexed | 2024-12-12T15:34:04Z |
| publishDate | 2022-06-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Ophthalmology Science |
| spelling | doaj.art-2e71409d77a341a8b6c30ab2a5e34f772022-12-22T00:20:03ZengElsevierOphthalmology Science2666-91452022-06-0122100126Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial IntelligenceAaron S. Coyner, PhD0Jimmy S. Chen, MD1Ken Chang, PhD2Praveer Singh, PhD3Susan Ostmo, MS4R. V. Paul Chan, MD, MSc5Michael F. Chiang, MD, MA6Jayashree Kalpathy-Cramer, PhD7J. Peter Campbell, MD, MPH8Department of Ophthalmology, Casey Eye Institute, Oregon Health & Science University, Portland, OregonDepartment of Ophthalmology, Casey Eye Institute, Oregon Health & Science University, Portland, Oregon; Department of Ophthalmology, Shiley Eye Institute, University of California, San Diego, San Diego, CaliforniaDepartment of Radiology, Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts; Center for Clinical Data Science, Massachusetts General Hospital and Boston Women’s Hospital, Boston, MassachusettsDepartment of Radiology, Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts; Center for Clinical Data Science, Massachusetts General Hospital and Boston Women’s Hospital, Boston, MassachusettsDepartment of Ophthalmology, Casey Eye Institute, Oregon Health & Science University, Portland, OregonDepartment of Ophthalmology and Visual Sciences, Eye and Ear Infirmary, University of Illinois, Chicago, IllinoisNational Eye Institute, National Institutes of Health, Bethesda, MarylandDepartment of Radiology, Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts; Center for Clinical Data Science, Massachusetts General Hospital and Boston Women’s Hospital, Boston, MassachusettsDepartment of Ophthalmology, Casey Eye Institute, Oregon Health & Science University, Portland, Oregon; Correspondence: J. Peter Campbell, MD, MPH, Department of Ophthalmology, Casey Eye Institute, Oregon Health & Science University, 545 SW Campus Drive, Portland, OR 97239.Purpose: Developing robust artificial intelligence (AI) models for medical image analysis requires large quantities of diverse, well-chosen data that can prove challenging to collect because of privacy concerns, disease rarity, or diagnostic label quality. Collecting image-based datasets for retinopathy of prematurity (ROP), a potentially blinding disease, suffers from these challenges. Progressively growing generative adversarial networks (PGANs) may help, because they can synthesize highly realistic images that may increase both the size and diversity of medical datasets. Design: Diagnostic validation study of convolutional neural networks (CNNs) for plus disease detection, a component of severe ROP, using synthetic data. Participants: Five thousand eight hundred forty-two retinal fundus images (RFIs) collected from 963 preterm infants. Methods: Retinal vessel maps (RVMs) were segmented from RFIs. PGANs were trained to synthesize RVMs with normal, pre-plus, or plus disease vasculature. Convolutional neural networks were trained, using real or synthetic RVMs, to detect plus disease from 2 real RVM test datasets. Main Outcome Measures: Features of real and synthetic RVMs were evaluated using uniform manifold approximation and projection (UMAP). Similarities were evaluated at the dataset and feature level using Fréchet inception distance and Euclidean distance, respectively. CNN performance was assessed via area under the receiver operating characteristic curve (AUC); AUCs were compared via bootstrapping and Delong’s test for correlated receiver operating characteristic curves. Confusion matrices were compared using McNemar’s chi-square test and Cohen’s κ value. Results: The CNN trained on synthetic RVMs showed a significantly higher AUC (0.971; P = 0.006 and P = 0.004) and classified plus disease more similarly to a set of 8 international experts (κ = 0.922) than the CNN trained on real RVMs (AUC = 0.934; κ = 0.701). Real and synthetic RVMs overlapped, by plus disease diagnosis, on the UMAP manifold, showing that synthetic images spanned the disease severity spectrum. Fréchet inception distance and Euclidean distances suggested that real and synthetic RVMs were more dissimilar to one another than real RVMs were to one another, further suggesting that synthetic RVMs were distinct from the training data with respect to privacy considerations. Conclusions: Synthetic datasets may be useful for training robust medical AI models. Furthermore, PGANs may be able to synthesize realistic data for use without protected health information concerns.http://www.sciencedirect.com/science/article/pii/S266691452200015XArtificial intelligenceDeep learningGenerative adversarial networkRetinopathy of prematurity |
| spellingShingle | Aaron S. Coyner, PhD Jimmy S. Chen, MD Ken Chang, PhD Praveer Singh, PhD Susan Ostmo, MS R. V. Paul Chan, MD, MSc Michael F. Chiang, MD, MA Jayashree Kalpathy-Cramer, PhD J. Peter Campbell, MD, MPH Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence Ophthalmology Science Artificial intelligence Deep learning Generative adversarial network Retinopathy of prematurity |
| title | Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence |
| title_full | Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence |
| title_fullStr | Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence |
| title_full_unstemmed | Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence |
| title_short | Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence |
| title_sort | synthetic medical images for robust privacy preserving training of artificial intelligence |
| topic | Artificial intelligence Deep learning Generative adversarial network Retinopathy of prematurity |
| url | http://www.sciencedirect.com/science/article/pii/S266691452200015X |
| work_keys_str_mv | AT aaronscoynerphd syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT jimmyschenmd syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT kenchangphd syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT praveersinghphd syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT susanostmoms syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT rvpaulchanmdmsc syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT michaelfchiangmdma syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT jayashreekalpathycramerphd syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence AT jpetercampbellmdmph syntheticmedicalimagesforrobustprivacypreservingtrainingofartificialintelligence |