Theoretical analysis and experimental validation of volume bias of soft Dice optimized segmentation maps in the context of inherent uncertainty

dc.contributor.authorBertels, Jeroen
dc.contributor.authorRobben, David
dc.contributor.authorVandermeulen, Dirk
dc.contributor.authorSuetens, Paul
dc.date.accessioned2022-11-09T10:33:51Z
dc.date.available2022-11-09T10:33:51Z
dc.date.issued2021-01
dc.description.abstractThe clinical interest is often to measure the volume of a structure, which is typically derived from a segmentation. In order to evaluate and compare segmentation methods, the similarity between a segmentation and a predefined ground truth is measured using popular discrete metrics, such as the Dice score. Recent segmentation methods use a differentiable surrogate metric, such as soft Dice, as part of the loss function during the learning phase. In this work, we first briefly describe how to derive volume estimates from a segmentation that is, potentially, inherently uncertain or ambiguous. This is followed by a theoretical analysis and an experimental validation linking the inherent uncertainty to common loss functions for training CNNs, namely cross-entropy and soft Dice. We find that, even though soft Dice optimization leads to an improved performance with respect to the Dice score and other measures, it may introduce a volume bias for tasks with high inherent uncertainty. These findings indicate some of the method’s clinical limitations and suggest doing a closer ad-hoc volume analysis with an optional re-calibration step.en_US
dc.description.departmentAnatomyen_US
dc.description.librarianhj2022en_US
dc.description.sponsorshipNEXIS (www.nexis-project.eu), a project that has received funding from the European Union’s Horizon 2020 Research and Innovations Programme and an innovation mandate of Flanders Innovation and Entrepreneurship (VLAIO).en_US
dc.description.urihttp://www.elsevier.com/locate/mediaen_US
dc.identifier.citation2021, 'Theoretical analysis and experimental validation of volume bias of soft Dice optimized segmentation maps in the context of inherent uncertainty', Medical Image Analysis, vol. 67, art. 101833, pp. 1-12, doi : 10.1016/j.media.2020.101833.en_US
dc.identifier.issn1361-8415
dc.identifier.other10.1016/j.media.2020.101833
dc.identifier.urihttps://repository.up.ac.za/handle/2263/88219
dc.language.isoenen_US
dc.publisherElsevieren_US
dc.rights© 2020 Elsevier B.V. All rights reserved. Notice : this is the author’s version of a work that was accepted for publication in Medical Image Analysis. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. A definitive version was subsequently published in Medical Image Analysis, vol. 67, art. 101833, pp. 1-12, 2021, doi : 10.1016/j.media.2020.101833.en_US
dc.subjectConvolutional neural network (CNN)en_US
dc.subjectSegmentationen_US
dc.subjectVolumeen_US
dc.subjectUncertaintyen_US
dc.subjectCross-entropyen_US
dc.subjectSoft diceen_US
dc.titleTheoretical analysis and experimental validation of volume bias of soft Dice optimized segmentation maps in the context of inherent uncertaintyen_US
dc.typePreprint Articleen_US

Files

Original bundle

Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
Bertels_Theoretical_2021.pdf
Size:
5.1 MB
Format:
Adobe Portable Document Format
Description:
Preprint Article

License bundle

Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
license.txt
Size:
1.75 KB
Format:
Item-specific license agreed upon to submission
Description: