Domain-adversarial neural network for improved generalization performance of gleason grade classification
(2020) Medical Imaging 2020: Digital Pathology In Progress in Biomedical Optics and Imaging - Proceedings of SPIE 11320.- Abstract
When training a deep learning model, the dataset used is of great importance to make sure that the model learns relevant features of the data and that it will be able to generalize to new data. However, it is typically difficult to produce a dataset without some bias toward any specific feature. Deep learning models used in histopathology have a tendency to overfit to the stain appearance of the training data - if the model is trained on data from one lab only, it will usually not be able to generalize to data from other labs. The standard technique to overcome this problem is to use color augmentation of the training data which, artificially, generates more variations for the network to learn. In this work we instead test the use of a... (More)
When training a deep learning model, the dataset used is of great importance to make sure that the model learns relevant features of the data and that it will be able to generalize to new data. However, it is typically difficult to produce a dataset without some bias toward any specific feature. Deep learning models used in histopathology have a tendency to overfit to the stain appearance of the training data - if the model is trained on data from one lab only, it will usually not be able to generalize to data from other labs. The standard technique to overcome this problem is to use color augmentation of the training data which, artificially, generates more variations for the network to learn. In this work we instead test the use of a so called domain-adversarial neural network, which is designed to prevent the model from being biased towards features that in reality are irrelevant such as the origin of an image. To test the technique, four datasets from different hospitals for Gleason grading of prostate cancer are used. We achieve state of the art results for these particular datasets, and furthermore for two of our three test datasets the approach outperforms the use of color augmentation.
(Less)
- author
- Arvidsson, Ida LU ; Overgaard, Niels Christian LU ; Krzyzanowska, Agnieszka LU ; Marginean, Felicia Elena LU ; Simoulis, Athanasios LU ; Bjartell, Anders LU ; Aström, Kalle LU and Heyden, Anders LU
- organization
-
- Mathematics (Faculty of Engineering)
- eSSENCE: The e-Science Collaboration
- ELLIIT: the Linköping-Lund initiative on IT and mobile communication
- Mathematical Imaging Group (research group)
- Partial differential equations (research group)
- LUCC: Lund University Cancer Centre
- Urological cancer, Malmö (research group)
- EpiHealth: Epidemiology for Health
- Stroke Imaging Research group (research group)
- publishing date
- 2020
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- Domain adversarial neural network, Generalization, Gleason grading
- host publication
- Medical Imaging 2020 : Digital Pathology - Digital Pathology
- series title
- Progress in Biomedical Optics and Imaging - Proceedings of SPIE
- editor
- Tomaszewski, John E. and Ward, Aaron D.
- volume
- 11320
- article number
- 1132016
- publisher
- SPIE
- conference name
- Medical Imaging 2020: Digital Pathology
- conference location
- Houston, United States
- conference dates
- 2020-02-19 - 2020-02-20
- external identifiers
-
- scopus:85120979234
- ISSN
- 1605-7422
- ISBN
- 9781510634077
- DOI
- 10.1117/12.2549011
- language
- English
- LU publication?
- yes
- id
- 845b99ac-cb9d-4784-8d0a-d3ede62e4ef6
- date added to LUP
- 2022-02-03 14:11:23
- date last changed
- 2024-04-09 00:18:03
@inproceedings{845b99ac-cb9d-4784-8d0a-d3ede62e4ef6, abstract = {{<p>When training a deep learning model, the dataset used is of great importance to make sure that the model learns relevant features of the data and that it will be able to generalize to new data. However, it is typically difficult to produce a dataset without some bias toward any specific feature. Deep learning models used in histopathology have a tendency to overfit to the stain appearance of the training data - if the model is trained on data from one lab only, it will usually not be able to generalize to data from other labs. The standard technique to overcome this problem is to use color augmentation of the training data which, artificially, generates more variations for the network to learn. In this work we instead test the use of a so called domain-adversarial neural network, which is designed to prevent the model from being biased towards features that in reality are irrelevant such as the origin of an image. To test the technique, four datasets from different hospitals for Gleason grading of prostate cancer are used. We achieve state of the art results for these particular datasets, and furthermore for two of our three test datasets the approach outperforms the use of color augmentation.</p>}}, author = {{Arvidsson, Ida and Overgaard, Niels Christian and Krzyzanowska, Agnieszka and Marginean, Felicia Elena and Simoulis, Athanasios and Bjartell, Anders and Aström, Kalle and Heyden, Anders}}, booktitle = {{Medical Imaging 2020 : Digital Pathology}}, editor = {{Tomaszewski, John E. and Ward, Aaron D.}}, isbn = {{9781510634077}}, issn = {{1605-7422}}, keywords = {{Domain adversarial neural network; Generalization; Gleason grading}}, language = {{eng}}, publisher = {{SPIE}}, series = {{Progress in Biomedical Optics and Imaging - Proceedings of SPIE}}, title = {{Domain-adversarial neural network for improved generalization performance of gleason grade classification}}, url = {{http://dx.doi.org/10.1117/12.2549011}}, doi = {{10.1117/12.2549011}}, volume = {{11320}}, year = {{2020}}, }