Comparison of different augmentation techniques for improved generalization performance for gleason grading
(2019) 16th IEEE International Symposium on Biomedical Imaging, ISBI 2019 p.923-927- Abstract
The fact that deep learning based algorithms used for digital pathology tend to overfit to the site of the training data is well-known. Since an algorithm that does not generalize is not very useful, we have in this work studied how different data augmentation techniques can reduce this problem but also how data from different sites can be normalized to each other. For both of these approaches we have used cycle generative adversarial networks (GAN); either to generate more examples to train on or to transform images from one site to another. Furthermore, we have investigated to what extent standard augmentation techniques improve the generalization performance. We performed experiments on four datasets with slides from prostate... (More)
The fact that deep learning based algorithms used for digital pathology tend to overfit to the site of the training data is well-known. Since an algorithm that does not generalize is not very useful, we have in this work studied how different data augmentation techniques can reduce this problem but also how data from different sites can be normalized to each other. For both of these approaches we have used cycle generative adversarial networks (GAN); either to generate more examples to train on or to transform images from one site to another. Furthermore, we have investigated to what extent standard augmentation techniques improve the generalization performance. We performed experiments on four datasets with slides from prostate biopsies, stained with HE, detailed annotated with Gleason grades. We obtained results similar to previous studies, with accuracies of 77% for Gleason grading for images from the same site as the training data and 59% for images from other sites. However, we also found out that the use of traditional augmentation techniques gave better performance compared to when using cycle GANs, either to augment the training data or to normalize the test data.
(Less)
- author
- Arvidsson, Ida LU ; Overgaard, Niels Christian LU ; Astrom, Kalle LU and Heyden, Anders LU
- organization
- publishing date
- 2019
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- Augmentation, Convolutional neural network, Cycle generative adversarial network, Deep learning, Digital pathology, Gleason grading, Prostate cancer
- host publication
- 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)
- article number
- 8759264
- pages
- 5 pages
- publisher
- IEEE Computer Society
- conference name
- 16th IEEE International Symposium on Biomedical Imaging, ISBI 2019
- conference location
- Venice, Italy
- conference dates
- 2019-04-08 - 2019-04-11
- external identifiers
-
- scopus:85073914841
- ISBN
- 9781538636411
- DOI
- 10.1109/ISBI.2019.8759264
- language
- English
- LU publication?
- yes
- id
- 32f41e4b-d2e7-4f3c-a45a-9e2aed57cb5d
- date added to LUP
- 2019-11-06 11:04:35
- date last changed
- 2023-12-04 03:43:17
@inproceedings{32f41e4b-d2e7-4f3c-a45a-9e2aed57cb5d, abstract = {{<p>The fact that deep learning based algorithms used for digital pathology tend to overfit to the site of the training data is well-known. Since an algorithm that does not generalize is not very useful, we have in this work studied how different data augmentation techniques can reduce this problem but also how data from different sites can be normalized to each other. For both of these approaches we have used cycle generative adversarial networks (GAN); either to generate more examples to train on or to transform images from one site to another. Furthermore, we have investigated to what extent standard augmentation techniques improve the generalization performance. We performed experiments on four datasets with slides from prostate biopsies, stained with HE, detailed annotated with Gleason grades. We obtained results similar to previous studies, with accuracies of 77% for Gleason grading for images from the same site as the training data and 59% for images from other sites. However, we also found out that the use of traditional augmentation techniques gave better performance compared to when using cycle GANs, either to augment the training data or to normalize the test data.</p>}}, author = {{Arvidsson, Ida and Overgaard, Niels Christian and Astrom, Kalle and Heyden, Anders}}, booktitle = {{2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)}}, isbn = {{9781538636411}}, keywords = {{Augmentation; Convolutional neural network; Cycle generative adversarial network; Deep learning; Digital pathology; Gleason grading; Prostate cancer}}, language = {{eng}}, pages = {{923--927}}, publisher = {{IEEE Computer Society}}, title = {{Comparison of different augmentation techniques for improved generalization performance for gleason grading}}, url = {{http://dx.doi.org/10.1109/ISBI.2019.8759264}}, doi = {{10.1109/ISBI.2019.8759264}}, year = {{2019}}, }