Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset

Hari Sreedhar; Guillaume P.R. Lajoinie; Charles Raffaelli; Hervé Delingette

doi:10.1007/978-3-031-58171-7_13

Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset

Hari Sreedhar^*
, Guillaume P.R. Lajoinie
, Charles Raffaelli
, Hervé Delingette

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

47 Downloads (Pure)

Abstract

Machine learning applications in ultrasound imaging are limited by access to ground-truth expert annotations, especially in specialized applications such as thyroid nodule evaluation. Active learning strategies seek to alleviate this concern by making more effective use of expert annotations; however, many proposed techniques do not adapt well to small-scale (i.e. a few hundred images) datasets. In this work, we test active learning strategies including an uncertainty-weighted selection approach with supervised and semi-supervised learning to evaluate the effectiveness of these tools for the prediction of nodule presence on a clinical ultrasound dataset. The results on this as well as two other medical image datasets suggest that even successful active learning strategies have limited clinical significance in terms of reducing annotation burden.

Original language	English
Title of host publication	Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings
Editors	Yuan Xue, Chen Chen, Chao Chen, Lianrui Zuo, Yihao Liu
Publisher	Springer
Pages	127-136
Number of pages	10
ISBN (Print)	9783031581700
DOIs	https://doi.org/10.1007/978-3-031-58171-7_13
Publication status	Published - 27 Apr 2024
Event	3rd International Workshop on Data Augmentation, Labeling, and Imperfections, DALI 2023 - Vancouver, Canada Duration: 12 Oct 2023 → 12 Oct 2023 Conference number: 3

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14379 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Workshop

Workshop	3rd International Workshop on Data Augmentation, Labeling, and Imperfections, DALI 2023
Abbreviated title	DALI 2023
Country/Territory	Canada
City	Vancouver
Period	12/10/23 → 12/10/23
Other	In conjunction with the 26th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2023

Keywords

2024 OA procedure
Thyroid cancer
Ultrasound imaging
Active learning

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1007/978-3-031-58171-7_13

978-3-031-58171-7_13Final published version, 850 KBLicence: Taverne

Cite this

Sreedhar, H., Lajoinie, G. P. R., Raffaelli, C., & Delingette, H. (2024). Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset. In Y. Xue, C. Chen, C. Chen, L. Zuo, & Y. Liu (Eds.), Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings (pp. 127-136). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14379 LNCS). Springer. https://doi.org/10.1007/978-3-031-58171-7_13

Sreedhar, Hari ; Lajoinie, Guillaume P.R. ; Raffaelli, Charles et al. / Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset. Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings. editor / Yuan Xue ; Chen Chen ; Chao Chen ; Lianrui Zuo ; Yihao Liu. Springer, 2024. pp. 127-136 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{9810ed63de2947599092492deea71731,

title = "Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset",

abstract = "Machine learning applications in ultrasound imaging are limited by access to ground-truth expert annotations, especially in specialized applications such as thyroid nodule evaluation. Active learning strategies seek to alleviate this concern by making more effective use of expert annotations; however, many proposed techniques do not adapt well to small-scale (i.e. a few hundred images) datasets. In this work, we test active learning strategies including an uncertainty-weighted selection approach with supervised and semi-supervised learning to evaluate the effectiveness of these tools for the prediction of nodule presence on a clinical ultrasound dataset. The results on this as well as two other medical image datasets suggest that even successful active learning strategies have limited clinical significance in terms of reducing annotation burden.",

keywords = "2024 OA procedure, Thyroid cancer, Ultrasound imaging, Active learning",

author = "Hari Sreedhar and Lajoinie, \{Guillaume P.R.\} and Charles Raffaelli and Herv{\'e} Delingette",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.; 3rd International Workshop on Data Augmentation, Labeling, and Imperfections, DALI 2023, DALI 2023 ; Conference date: 12-10-2023 Through 12-10-2023",

year = "2024",

month = apr,

day = "27",

doi = "10.1007/978-3-031-58171-7\_13",

language = "English",

isbn = "9783031581700",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "127--136",

editor = "Yuan Xue and Chen Chen and Chao Chen and Lianrui Zuo and Yihao Liu",

booktitle = "Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings",

address = "Germany",

}

Sreedhar, H, Lajoinie, GPR, Raffaelli, C & Delingette, H 2024, Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset. in Y Xue, C Chen, C Chen, L Zuo & Y Liu (eds), Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14379 LNCS, Springer, pp. 127-136, 3rd International Workshop on Data Augmentation, Labeling, and Imperfections, DALI 2023, Vancouver, British Columbia, Canada, 12/10/23. https://doi.org/10.1007/978-3-031-58171-7_13

Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset. / Sreedhar, Hari; Lajoinie, Guillaume P.R.; Raffaelli, Charles et al.
Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings. ed. / Yuan Xue; Chen Chen; Chao Chen; Lianrui Zuo; Yihao Liu. Springer, 2024. p. 127-136 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14379 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset

AU - Sreedhar, Hari

AU - Lajoinie, Guillaume P.R.

AU - Raffaelli, Charles

AU - Delingette, Hervé

N1 - Conference code: 3

PY - 2024/4/27

Y1 - 2024/4/27

N2 - Machine learning applications in ultrasound imaging are limited by access to ground-truth expert annotations, especially in specialized applications such as thyroid nodule evaluation. Active learning strategies seek to alleviate this concern by making more effective use of expert annotations; however, many proposed techniques do not adapt well to small-scale (i.e. a few hundred images) datasets. In this work, we test active learning strategies including an uncertainty-weighted selection approach with supervised and semi-supervised learning to evaluate the effectiveness of these tools for the prediction of nodule presence on a clinical ultrasound dataset. The results on this as well as two other medical image datasets suggest that even successful active learning strategies have limited clinical significance in terms of reducing annotation burden.

AB - Machine learning applications in ultrasound imaging are limited by access to ground-truth expert annotations, especially in specialized applications such as thyroid nodule evaluation. Active learning strategies seek to alleviate this concern by making more effective use of expert annotations; however, many proposed techniques do not adapt well to small-scale (i.e. a few hundred images) datasets. In this work, we test active learning strategies including an uncertainty-weighted selection approach with supervised and semi-supervised learning to evaluate the effectiveness of these tools for the prediction of nodule presence on a clinical ultrasound dataset. The results on this as well as two other medical image datasets suggest that even successful active learning strategies have limited clinical significance in terms of reducing annotation burden.

KW - 2024 OA procedure

KW - Thyroid cancer

KW - Ultrasound imaging

KW - Active learning

UR - https://www.scopus.com/pages/publications/85192908499

U2 - 10.1007/978-3-031-58171-7_13

DO - 10.1007/978-3-031-58171-7_13

M3 - Conference contribution

AN - SCOPUS:85192908499

SN - 9783031581700

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 127

EP - 136

BT - Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings

A2 - Xue, Yuan

A2 - Chen, Chen

A2 - Chen, Chao

A2 - Zuo, Lianrui

A2 - Liu, Yihao

PB - Springer

T2 - 3rd International Workshop on Data Augmentation, Labeling, and Imperfections, DALI 2023

Y2 - 12 October 2023 through 12 October 2023

ER -

Sreedhar H, Lajoinie GPR, Raffaelli C, Delingette H. Active Learning Strategies on a Real-World Thyroid Ultrasound Dataset. In Xue Y, Chen C, Chen C, Zuo L, Liu Y, editors, Data Augmentation, Labelling, and Imperfections - 3rd MICCAI Workshop, DALI 2023 Held in Conjunction with MICCAI 2023, Proceedings. Springer. 2024. p. 127-136. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-58171-7_13