Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations

TY - JOUR

T1 - Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations

AU - Sudre, Carole H.

AU - Li, Wenqi

AU - Vercauteren, Tom

AU - Ourselin, Sebastien

AU - Cardoso, M. Jorge

PY - 2017/9/9

Y1 - 2017/9/9

N2 - Deep-learning has proved in recent years to be a powerful tool for image analysis and is now widely used to segment both 2D and 3D medical images. Deep-learning segmentation frameworks rely not only on the choice of network architecture but also on the choice of loss function. When the segmentation process targets rare observations, a severe class imbalance is likely to occur between candidate labels, thus resulting in sub-optimal performance. In order to mitigate this issue, strategies such as the weighted cross-entropy function, the sensitivity function or the Dice loss function, have been proposed. In this work, we investigate the behavior of these loss functions and their sensitivity to learning rate tuning in the presence of different rates of label imbalance across 2D and 3D segmentation tasks. We also propose to use the class re-balancing properties of the Generalized Dice overlap, a known metric for segmentation assessment, as a robust and accurate deep-learning loss function for unbalanced tasks.

AB - Deep-learning has proved in recent years to be a powerful tool for image analysis and is now widely used to segment both 2D and 3D medical images. Deep-learning segmentation frameworks rely not only on the choice of network architecture but also on the choice of loss function. When the segmentation process targets rare observations, a severe class imbalance is likely to occur between candidate labels, thus resulting in sub-optimal performance. In order to mitigate this issue, strategies such as the weighted cross-entropy function, the sensitivity function or the Dice loss function, have been proposed. In this work, we investigate the behavior of these loss functions and their sensitivity to learning rate tuning in the presence of different rates of label imbalance across 2D and 3D segmentation tasks. We also propose to use the class re-balancing properties of the Generalized Dice overlap, a known metric for segmentation assessment, as a robust and accurate deep-learning loss function for unbalanced tasks.

UR - http://www.scopus.com/inward/record.url?scp=85029792269&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-67558-9_28

DO - 10.1007/978-3-319-67558-9_28

M3 - Conference paper

AN - SCOPUS:85029792269

SN - 0302-9743

VL - 10553 LNCS

SP - 240

EP - 248

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

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

T2 - 3rd International Workshop on Deep Learning in Medical Image Analysis, DLMIA 2017 and 7th International Workshop on Multimodal Learning for Clinical Decision Support, ML-CDS 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017

Y2 - 14 September 2017 through 14 September 2017

ER -