Can Generative Adversarial Networks help to overcome the limited data problem in segmentation?
Purpose: For image translational tasks, the application of deep learning methods showed that Generative Adversarial Network (GAN) architectures outperform the traditional U-Net networks, when using the same training data size. This study investigates whether this performance boost can also be expect...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2022-08-01
|
| Series: | Zeitschrift für Medizinische Physik |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S0939388921001124 |
| _version_ | 1797385297538842624 |
|---|---|
| author | Gerd Heilemann Mark Matthewman Peter Kuess Gregor Goldner Joachim Widder Dietmar Georg Lukas Zimmermann |
| author_facet | Gerd Heilemann Mark Matthewman Peter Kuess Gregor Goldner Joachim Widder Dietmar Georg Lukas Zimmermann |
| author_sort | Gerd Heilemann |
| collection | DOAJ |
| description | Purpose: For image translational tasks, the application of deep learning methods showed that Generative Adversarial Network (GAN) architectures outperform the traditional U-Net networks, when using the same training data size. This study investigates whether this performance boost can also be expected for segmentation tasks with small training dataset size. Materials/Methods: Two models were trained on varying training dataset sizes ranging from 1—100 patients: a) U-Net and b) U-Net with patch discriminator (conditional GAN). The performance of both models to segment the male pelvis on CT-data was evaluated (Dice similarity coefficient, Hausdorff) with respect to training data size. Results: No significant differences were observed between the U-Net and cGAN when the models were trained with the same training sizes up to 100 patients. The training dataset size had a significant impact on the models’ performances, with vast improvements when increasing dataset sizes from 1 to 20 patients. Conclusion: When introducing GANs for the segmentation task no significant performance boost was observed in our experiments, even in segmentation models developed on small datasets. |
| first_indexed | 2024-03-08T21:52:05Z |
| format | Article |
| id | doaj.art-9e283f98eaf04e05a899a51a275ad360 |
| institution | Directory Open Access Journal |
| issn | 0939-3889 |
| language | English |
| last_indexed | 2024-03-08T21:52:05Z |
| publishDate | 2022-08-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Zeitschrift für Medizinische Physik |
| spelling | doaj.art-9e283f98eaf04e05a899a51a275ad3602023-12-20T07:33:12ZengElsevierZeitschrift für Medizinische Physik0939-38892022-08-01323361368Can Generative Adversarial Networks help to overcome the limited data problem in segmentation?Gerd Heilemann0Mark Matthewman1Peter Kuess2Gregor Goldner3Joachim Widder4Dietmar Georg5Lukas Zimmermann6Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria; Comprehensive Cancer Center, Medical University of Vienna, Vienna, Austria; Corresponding author: Gerd Heilemann, Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria.Technical University of Vienna, Vienna, AustriaDepartment of Radiation Oncology, Medical University of Vienna, Vienna, Austria; Comprehensive Cancer Center, Medical University of Vienna, Vienna, AustriaDepartment of Radiation Oncology, Medical University of Vienna, Vienna, Austria; Comprehensive Cancer Center, Medical University of Vienna, Vienna, AustriaDepartment of Radiation Oncology, Medical University of Vienna, Vienna, Austria; Comprehensive Cancer Center, Medical University of Vienna, Vienna, AustriaDepartment of Radiation Oncology, Medical University of Vienna, Vienna, Austria; Comprehensive Cancer Center, Medical University of Vienna, Vienna, AustriaDepartment of Radiation Oncology, Medical University of Vienna, Vienna, Austria; Competence Center for Preclinical Imaging and Biomedical Engineering, University of Applied Sciences Wiener Neustadt, Austria; Faculty of Engineering, University of Applied Sciences Wiener Neustadt, AustriaPurpose: For image translational tasks, the application of deep learning methods showed that Generative Adversarial Network (GAN) architectures outperform the traditional U-Net networks, when using the same training data size. This study investigates whether this performance boost can also be expected for segmentation tasks with small training dataset size. Materials/Methods: Two models were trained on varying training dataset sizes ranging from 1—100 patients: a) U-Net and b) U-Net with patch discriminator (conditional GAN). The performance of both models to segment the male pelvis on CT-data was evaluated (Dice similarity coefficient, Hausdorff) with respect to training data size. Results: No significant differences were observed between the U-Net and cGAN when the models were trained with the same training sizes up to 100 patients. The training dataset size had a significant impact on the models’ performances, with vast improvements when increasing dataset sizes from 1 to 20 patients. Conclusion: When introducing GANs for the segmentation task no significant performance boost was observed in our experiments, even in segmentation models developed on small datasets.http://www.sciencedirect.com/science/article/pii/S0939388921001124Automatic segmentationDeep learningProstate cancerGenerative adversarial networks |
| spellingShingle | Gerd Heilemann Mark Matthewman Peter Kuess Gregor Goldner Joachim Widder Dietmar Georg Lukas Zimmermann Can Generative Adversarial Networks help to overcome the limited data problem in segmentation? Zeitschrift für Medizinische Physik Automatic segmentation Deep learning Prostate cancer Generative adversarial networks |
| title | Can Generative Adversarial Networks help to overcome the limited data problem in segmentation? |
| title_full | Can Generative Adversarial Networks help to overcome the limited data problem in segmentation? |
| title_fullStr | Can Generative Adversarial Networks help to overcome the limited data problem in segmentation? |
| title_full_unstemmed | Can Generative Adversarial Networks help to overcome the limited data problem in segmentation? |
| title_short | Can Generative Adversarial Networks help to overcome the limited data problem in segmentation? |
| title_sort | can generative adversarial networks help to overcome the limited data problem in segmentation |
| topic | Automatic segmentation Deep learning Prostate cancer Generative adversarial networks |
| url | http://www.sciencedirect.com/science/article/pii/S0939388921001124 |
| work_keys_str_mv | AT gerdheilemann cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation AT markmatthewman cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation AT peterkuess cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation AT gregorgoldner cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation AT joachimwidder cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation AT dietmargeorg cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation AT lukaszimmermann cangenerativeadversarialnetworkshelptoovercomethelimiteddataprobleminsegmentation |