Dice Semimetric Losses: Optimizing the Dice Score with Soft Labels
Authors:
Preprint
Abstract
The soft Dice loss (SDL) has taken a pivotal role in many automated segmentation pipelines in the medical imaging community. Over the last years, some reasons behind its superior functioning have been uncovered and further optimizations have been explored. However, there is currently no implementation that supports its direct use in settings with soft labels. Hence, a synergy between the use of SDL and research leveraging the use of soft labels, also in the context of model calibration, is still missing. In this work, we introduce Dice semimetric losses (DMLs), which (i) are by design identical to SDL in a standard setting with hard labels, but (ii) can be used in settings with soft labels. Our experiments on the public QUBIQ, LiTS and KiTS benchmarks confirm the potential synergy of DMLs with soft labels (e.g. averaging, label smoothing, and knowledge distillation) over hard labels (e.g. majority voting and random selection). As a result, we obtain superior Dice scores and model calibration, which supports the wider adoption of DMLs in practice. Code is available at \href{https://github.com/zifuwanggg/JDTLosses}{https://github.com/zifuwanggg/JDTLosses}.Keywords
All Related Versions
This publication has 26 references indexed in Scilit:
- The Liver Tumor Segmentation Benchmark (LiTS)Medical Image Analysis, 2022
- On the Relationship Between Calibrated Predictors and Unbiased Volume Estimation2021
- SoftSeg: Advantages of soft versus binary training for image segmentationMedical Image Analysis, 2021
- Spatially Varying Label Smoothing: Capturing Uncertainty from Expert Annotations2021
- Theoretical analysis and experimental validation of volume bias of soft Dice optimized segmentation maps in the context of inherent uncertaintyMedical Image Analysis, 2021
- The state of the art in kidney and kidney tumor segmentation in contrast-enhanced CT imaging: Results of the KiTS19 challengeMedical Image Analysis, 2020
- Improving Uncertainty Estimation in Convolutional Neural Networks Using Inter-rater Agreement2019
- Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations2017
- Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks2017
- U-Net: Convolutional Networks for Biomedical Image Segmentation2015