Transductive Image Segmentation : Self-training and Effect of Uncertainty Estimation
(2021) 3rd MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2021, and the 1st MICCAI Workshop on Affordable Healthcare and AI for Resource Diverse Global Health, FAIR 2021, held in conjunction with 24th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2021 In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12968 LNCS. p.79-89- Abstract
Semi-supervised learning (SSL) uses unlabeled data during training to learn better models. Previous studies on SSL for medical image segmentation focused mostly on improving model generalization to unseen data. In some applications, however, our primary interest is not generalization but to obtain optimal predictions on a specific unlabeled database that is fully available during model development. Examples include population studies for extracting imaging phenotypes. This work investigates an often overlooked aspect of SSL, transduction. It focuses on the quality of predictions made on the unlabeled data of interest when they are included for optimization during training, rather than improving generalization. We focus on the... (More)
Semi-supervised learning (SSL) uses unlabeled data during training to learn better models. Previous studies on SSL for medical image segmentation focused mostly on improving model generalization to unseen data. In some applications, however, our primary interest is not generalization but to obtain optimal predictions on a specific unlabeled database that is fully available during model development. Examples include population studies for extracting imaging phenotypes. This work investigates an often overlooked aspect of SSL, transduction. It focuses on the quality of predictions made on the unlabeled data of interest when they are included for optimization during training, rather than improving generalization. We focus on the self-training framework and explore its potential for transduction. We analyze it through the lens of Information Gain and reveal that learning benefits from the use of calibrated or under-confident models. Our extensive experiments on a large MRI database for multi-class segmentation of traumatic brain lesions shows promising results when comparing transductive with inductive predictions. We believe this study will inspire further research on transductive learning, a well-suited paradigm for medical image analysis.
(Less)
- author
- organization
- publishing date
- 2021
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- Domain Adaptation and Representation Transfer, and Affordable Healthcare and AI for Resource Diverse Global Health - 3rd MICCAI Workshop, DART 2021, and 1st MICCAI Workshop, FAIR 2021, Held in Conjunction with MICCAI 2021, Proceedings
- series title
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- editor
- Albarqouni, Shadi ; Cardoso, M. Jorge ; Dou, Qi ; Kamnitsas, Konstantinos ; Khanal, Bishesh ; Rekik, Islem ; Rieke, Nicola ; Sheet, Debdoot ; Tsaftaris, Sotirios ; Xu, Daguang and Xu, Ziyue
- volume
- 12968 LNCS
- pages
- 11 pages
- publisher
- Springer Science and Business Media B.V.
- conference name
- 3rd MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2021, and the 1st MICCAI Workshop on Affordable Healthcare and AI for Resource Diverse Global Health, FAIR 2021, held in conjunction with 24th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2021
- conference location
- Virtual, Online
- conference dates
- 2021-09-27 - 2021-10-01
- external identifiers
-
- scopus:85116422173
- ISSN
- 0302-9743
- 1611-3349
- ISBN
- 9783030877217
- DOI
- 10.1007/978-3-030-87722-4_8
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2021, Springer Nature Switzerland AG.
- id
- a30cb5e1-7eea-4c3a-80aa-e53c5f0ab701
- date added to LUP
- 2021-10-25 14:39:04
- date last changed
- 2024-03-23 12:20:35
@inproceedings{a30cb5e1-7eea-4c3a-80aa-e53c5f0ab701,
abstract = {{<p>Semi-supervised learning (SSL) uses unlabeled data during training to learn better models. Previous studies on SSL for medical image segmentation focused mostly on improving model generalization to unseen data. In some applications, however, our primary interest is not generalization but to obtain optimal predictions on a specific unlabeled database that is fully available during model development. Examples include population studies for extracting imaging phenotypes. This work investigates an often overlooked aspect of SSL, transduction. It focuses on the quality of predictions made on the unlabeled data of interest when they are included for optimization during training, rather than improving generalization. We focus on the self-training framework and explore its potential for transduction. We analyze it through the lens of Information Gain and reveal that learning benefits from the use of calibrated or under-confident models. Our extensive experiments on a large MRI database for multi-class segmentation of traumatic brain lesions shows promising results when comparing transductive with inductive predictions. We believe this study will inspire further research on transductive learning, a well-suited paradigm for medical image analysis.</p>}},
author = {{Kamnitsas, Konstantinos and Winzeck, Stefan and Kornaropoulos, Evgenios N. and Whitehouse, Daniel and Englman, Cameron and Phyu, Poe and Pao, Norman and Menon, David K. and Rueckert, Daniel and Das, Tilak and Newcombe, Virginia F.J. and Glocker, Ben}},
booktitle = {{Domain Adaptation and Representation Transfer, and Affordable Healthcare and AI for Resource Diverse Global Health - 3rd MICCAI Workshop, DART 2021, and 1st MICCAI Workshop, FAIR 2021, Held in Conjunction with MICCAI 2021, Proceedings}},
editor = {{Albarqouni, Shadi and Cardoso, M. Jorge and Dou, Qi and Kamnitsas, Konstantinos and Khanal, Bishesh and Rekik, Islem and Rieke, Nicola and Sheet, Debdoot and Tsaftaris, Sotirios and Xu, Daguang and Xu, Ziyue}},
isbn = {{9783030877217}},
issn = {{0302-9743}},
language = {{eng}},
pages = {{79--89}},
publisher = {{Springer Science and Business Media B.V.}},
series = {{Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}},
title = {{Transductive Image Segmentation : Self-training and Effect of Uncertainty Estimation}},
url = {{http://dx.doi.org/10.1007/978-3-030-87722-4_8}},
doi = {{10.1007/978-3-030-87722-4_8}},
volume = {{12968 LNCS}},
year = {{2021}},
}