Unsupervised GAN epoch selection for biomedical data synthesis
Supervised Neural Networks are used for segmentation in many biological and biomedical applications. To omit the time-consuming and tiring process of manual labeling, unsupervised Generative Adversarial Networks (GANs) can be used to synthesize labeled data. However, the training of GANs requires ex...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
De Gruyter
2023-09-01
|
| Series: | Current Directions in Biomedical Engineering |
| Subjects: | |
| Online Access: | https://doi.org/10.1515/cdbme-2023-1117 |
| _version_ | 1797646307251191808 |
|---|---|
| author | Böhland Moritz Bruch Roman Löffler Katharina Reischl Markus |
| author_facet | Böhland Moritz Bruch Roman Löffler Katharina Reischl Markus |
| author_sort | Böhland Moritz |
| collection | DOAJ |
| description | Supervised Neural Networks are used for segmentation in many biological and biomedical applications. To omit the time-consuming and tiring process of manual labeling, unsupervised Generative Adversarial Networks (GANs) can be used to synthesize labeled data. However, the training of GANs requires extensive computation and is often unstable. Due to the lack of established stopping criteria, GANs are usually trained multiple times for a heuristically fixed number of epochs. Early stopping and epoch selection can lead to better synthetic datasets resulting in higher downstream segmentation quality on biological or medical data. This article examines whether the Frechet Inception Distance (FID), the Kernel Inception Distance (KID), or the WeightWatcher tool can be used for early stopping or epoch selection of unsupervised GANs. The experiments show that the last trained GAN epoch is not necessarily the best one to synthesize downstream segmentation data. On complex datasets, FID and KID correlate with the downstream segmentation quality, and both can be used for epoch selection. |
| first_indexed | 2024-03-11T15:00:45Z |
| format | Article |
| id | doaj.art-62e4171019184ebd95160caf197d8286 |
| institution | Directory Open Access Journal |
| issn | 2364-5504 |
| language | English |
| last_indexed | 2024-03-11T15:00:45Z |
| publishDate | 2023-09-01 |
| publisher | De Gruyter |
| record_format | Article |
| series | Current Directions in Biomedical Engineering |
| spelling | doaj.art-62e4171019184ebd95160caf197d82862023-10-30T07:58:12ZengDe GruyterCurrent Directions in Biomedical Engineering2364-55042023-09-019146747010.1515/cdbme-2023-1117Unsupervised GAN epoch selection for biomedical data synthesisBöhland Moritz0Bruch Roman1Löffler Katharina2Reischl Markus3Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Hermann-von-Helmholtz-Platz 1, 76344Eggenstein-Leopoldshafen, GermanyInstitute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Hermann-von-Helmholtz-Platz 1, 76344Eggenstein-Leopoldshafen, GermanyInstitute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Hermann-von-Helmholtz-Platz 1, 76344Eggenstein-Leopoldshafen, GermanyInstitute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Hermann-von-Helmholtz-Platz 1, 76344Eggenstein-Leopoldshafen, GermanySupervised Neural Networks are used for segmentation in many biological and biomedical applications. To omit the time-consuming and tiring process of manual labeling, unsupervised Generative Adversarial Networks (GANs) can be used to synthesize labeled data. However, the training of GANs requires extensive computation and is often unstable. Due to the lack of established stopping criteria, GANs are usually trained multiple times for a heuristically fixed number of epochs. Early stopping and epoch selection can lead to better synthetic datasets resulting in higher downstream segmentation quality on biological or medical data. This article examines whether the Frechet Inception Distance (FID), the Kernel Inception Distance (KID), or the WeightWatcher tool can be used for early stopping or epoch selection of unsupervised GANs. The experiments show that the last trained GAN epoch is not necessarily the best one to synthesize downstream segmentation data. On complex datasets, FID and KID correlate with the downstream segmentation quality, and both can be used for epoch selection.https://doi.org/10.1515/cdbme-2023-1117generative adversarial networkdata synthesissegmentationcomputer vision |
| spellingShingle | Böhland Moritz Bruch Roman Löffler Katharina Reischl Markus Unsupervised GAN epoch selection for biomedical data synthesis Current Directions in Biomedical Engineering generative adversarial network data synthesis segmentation computer vision |
| title | Unsupervised GAN epoch selection for biomedical data synthesis |
| title_full | Unsupervised GAN epoch selection for biomedical data synthesis |
| title_fullStr | Unsupervised GAN epoch selection for biomedical data synthesis |
| title_full_unstemmed | Unsupervised GAN epoch selection for biomedical data synthesis |
| title_short | Unsupervised GAN epoch selection for biomedical data synthesis |
| title_sort | unsupervised gan epoch selection for biomedical data synthesis |
| topic | generative adversarial network data synthesis segmentation computer vision |
| url | https://doi.org/10.1515/cdbme-2023-1117 |
| work_keys_str_mv | AT bohlandmoritz unsupervisedganepochselectionforbiomedicaldatasynthesis AT bruchroman unsupervisedganepochselectionforbiomedicaldatasynthesis AT lofflerkatharina unsupervisedganepochselectionforbiomedicaldatasynthesis AT reischlmarkus unsupervisedganepochselectionforbiomedicaldatasynthesis |