Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps

Christian Payer; Martin Urschler; Horst Bischof; Darko Stern

doi:10.1007/978-3-030-60365-6_5

Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps

Christian Payer^*, Martin Urschler, Horst Bischof, Darko Stern

^*Korrespondierende/r Autor/-in für diese Arbeit

Institute of Visual Computing (7100)

Publikation: Beitrag in Buch/Bericht/Konferenzband › Beitrag in einem Konferenzband › Begutachtung

Abstract

In landmark localization, due to ambiguities in defining their exact position, landmark annotations may suffer from both large inter- and intra-observer variabilites, which result in uncertain annotations. Therefore, predicting a single coordinate for a landmark is not sufficient for modeling the distribution of possible landmark locations. We propose to learn the Gaussian covariances of target heatmaps, such that covariances for pointed heatmaps correspond to more certain landmarks and covariances for flat heatmaps to more uncertain or ambiguous landmarks. By fitting Gaussian functions to the predicted heatmaps, our method is able to obtain landmark location distributions, which model location uncertainties. We show on a dataset of left hand radiographs and on a dataset of lateral cephalograms that the predicted uncertainties correlate with the landmark error, as well as inter-observer variabilities

Originalsprache	englisch
Titel	Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis
Untertitel	Second International Workshop, UNSURE 2020, and Third International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings
Erscheinungsort	Cham
Herausgeber (Verlag)	Springer
Seiten	42-51
Seitenumfang	10
ISBN (elektronisch)	978-3-030-60365-6
ISBN (Print)	978-3-030-60364-9
DOIs	https://doi.org/10.1007/978-3-030-60365-6_5
Publikationsstatus	Veröffentlicht - 8 Okt. 2020
Veranstaltung	2nd International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging: UNSURE 2020 - Virtual, Lima, Peru Dauer: 8 Okt. 2020 → …

Publikationsreihe

Name	Lecture Notes in Computer Science
Band	12443

Workshop

Workshop	2nd International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging
Land/Gebiet	Peru
Ort	Virtual, Lima
Zeitraum	8/10/20 → …

ASJC Scopus subject areas

Theoretische Informatik
Allgemeine Computerwissenschaft

Zugriff auf Dokument

10.1007/978-3-030-60365-6_5

Andere Dateien und Links

http://www.scopus.com/inward/record.url?scp=85093069268&partnerID=8YFLogxK

Uncertainty Estimation in Landmark Localization based on Gaussian Heatmaps
Payer, C. (Redner/in)
8 Okt. 2020
Aktivität: Vortrag oder Präsentation › Vortrag bei Workshop, Seminar oder Kurs › Science to science

Dieses zitieren

Payer, C., Urschler, M., Bischof, H., & Stern, D. (2020). Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps. in Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis: Second International Workshop, UNSURE 2020, and Third International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings (S. 42-51). ( Lecture Notes in Computer Science; Band 12443). Springer. https://doi.org/10.1007/978-3-030-60365-6_5

Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps. / Payer, Christian; Urschler, Martin; Bischof, Horst et al.
Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis: Second International Workshop, UNSURE 2020, and Third International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. Cham: Springer, 2020. S. 42-51 ( Lecture Notes in Computer Science; Band 12443).

Publikation: Beitrag in Buch/Bericht/Konferenzband › Beitrag in einem Konferenzband › Begutachtung

Payer, C, Urschler, M, Bischof, H & Stern, D 2020, Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps. in Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis: Second International Workshop, UNSURE 2020, and Third International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. Lecture Notes in Computer Science, Bd. 12443, Springer, Cham, S. 42-51, 2nd International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, Virtual, Lima, Peru, 8/10/20. https://doi.org/10.1007/978-3-030-60365-6_5

Payer C, Urschler M, Bischof H, Stern D. Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps. in Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis: Second International Workshop, UNSURE 2020, and Third International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. Cham: Springer. 2020. S. 42-51. ( Lecture Notes in Computer Science). doi: 10.1007/978-3-030-60365-6_5

Payer, Christian ; Urschler, Martin ; Bischof, Horst et al. / Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps. Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis: Second International Workshop, UNSURE 2020, and Third International Workshop, GRAIL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. Cham : Springer, 2020. S. 42-51 ( Lecture Notes in Computer Science).

@inproceedings{8dc259ad684143b3b1e444b72dd94063,

title = "Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps",

abstract = "In landmark localization, due to ambiguities in defining their exact position, landmark annotations may suffer from both large inter- and intra-observer variabilites, which result in uncertain annotations. Therefore, predicting a single coordinate for a landmark is not sufficient for modeling the distribution of possible landmark locations. We propose to learn the Gaussian covariances of target heatmaps, such that covariances for pointed heatmaps correspond to more certain landmarks and covariances for flat heatmaps to more uncertain or ambiguous landmarks. By fitting Gaussian functions to the predicted heatmaps, our method is able to obtain landmark location distributions, which model location uncertainties. We show on a dataset of left hand radiographs and on a dataset of lateral cephalograms that the predicted uncertainties correlate with the landmark error, as well as inter-observer variabilities",

keywords = "Landmark localization, Uncertainty estimation",

author = "Christian Payer and Martin Urschler and Horst Bischof and Darko Stern",

year = "2020",

month = oct,

day = "8",

doi = "10.1007/978-3-030-60365-6_5",

language = "English",

isbn = "978-3-030-60364-9",

series = " Lecture Notes in Computer Science",

publisher = "Springer",

pages = "42--51",

booktitle = "Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis",

note = "2nd International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging : UNSURE 2020 ; Conference date: 08-10-2020",

}

TY - GEN

T1 - Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps

AU - Payer, Christian

AU - Urschler, Martin

AU - Bischof, Horst

AU - Stern, Darko

PY - 2020/10/8

Y1 - 2020/10/8

N2 - In landmark localization, due to ambiguities in defining their exact position, landmark annotations may suffer from both large inter- and intra-observer variabilites, which result in uncertain annotations. Therefore, predicting a single coordinate for a landmark is not sufficient for modeling the distribution of possible landmark locations. We propose to learn the Gaussian covariances of target heatmaps, such that covariances for pointed heatmaps correspond to more certain landmarks and covariances for flat heatmaps to more uncertain or ambiguous landmarks. By fitting Gaussian functions to the predicted heatmaps, our method is able to obtain landmark location distributions, which model location uncertainties. We show on a dataset of left hand radiographs and on a dataset of lateral cephalograms that the predicted uncertainties correlate with the landmark error, as well as inter-observer variabilities

AB - In landmark localization, due to ambiguities in defining their exact position, landmark annotations may suffer from both large inter- and intra-observer variabilites, which result in uncertain annotations. Therefore, predicting a single coordinate for a landmark is not sufficient for modeling the distribution of possible landmark locations. We propose to learn the Gaussian covariances of target heatmaps, such that covariances for pointed heatmaps correspond to more certain landmarks and covariances for flat heatmaps to more uncertain or ambiguous landmarks. By fitting Gaussian functions to the predicted heatmaps, our method is able to obtain landmark location distributions, which model location uncertainties. We show on a dataset of left hand radiographs and on a dataset of lateral cephalograms that the predicted uncertainties correlate with the landmark error, as well as inter-observer variabilities

KW - Landmark localization

KW - Uncertainty estimation

UR - http://www.scopus.com/inward/record.url?scp=85093069268&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-60365-6_5

DO - 10.1007/978-3-030-60365-6_5

M3 - Conference paper

SN - 978-3-030-60364-9

T3 - Lecture Notes in Computer Science

SP - 42

EP - 51

BT - Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Graphs in Biomedical Image Analysis

PB - Springer

CY - Cham

T2 - 2nd International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging

Y2 - 8 October 2020

ER -

Uncertainty Estimation in Landmark Localization Based on Gaussian Heatmaps

Abstract

Publikationsreihe

Workshop

ASJC Scopus subject areas

Zugriff auf Dokument

Andere Dateien und Links

Fingerprint

Aktivitäten

Uncertainty Estimation in Landmark Localization based on Gaussian Heatmaps

Dieses zitieren