Optimizing the Dice Score and Jaccard Index for Medical Image Segmentation: Theory and Practice

@inproceedings{Bertels2019OptimizingTD,
  title={Optimizing the Dice Score and Jaccard Index for Medical Image Segmentation: Theory and Practice},
  author={J. Bertels and Tom Eelbode and Maxim Berman and Dirk Vandermeulen and Frederik Maes and Raf Bisschops and Matthew B. Blaschko},
  booktitle={MICCAI},
  year={2019}
}
The Dice score and Jaccard index are commonly used metrics for the evaluation of segmentation tasks in medical imaging. Convolutional neural networks trained for image segmentation tasks are usually optimized for (weighted) cross-entropy. This introduces an adverse discrepancy between the learning optimization objective (the loss) and the end target metric. Recent works in computer vision have proposed soft surrogates to alleviate this discrepancy and directly optimize the desired metric… 

Optimization for Medical Image Segmentation: Theory and Practice When Evaluating With Dice Score or Jaccard Index

It is confirmed that metric-sensitive losses are superior to cross-entropy based loss functions in case of evaluation with Dice Score or Jaccard Index in a multi-class setting, and across different object sizes and foreground/background ratios.

On the Optimal Combination of Cross-Entropy and Soft Dice Losses for Lesion Segmentation with Out-of-Distribution Robustness

The results are surprising: CE-Dice loss combinations that excel in segmenting in-distribution images have a poor performance when dealing with OoD data, which leads the adoption of the CE loss for this kind of problems, due to its robustness and ability to generalize to OoD samples.

DOMINO: Domain-Aware Model Calibration in Medical Image Segmentation

The experiments demonstrate that the DOMINO-calibrated deep neural networks outperform non-Calibrated models and state-of-the-art morphometric methods in head image segmentation and show the importance of semantic ties between class labels in building confidence in deep learning models.

Region-wise Loss for Biomedical Image Segmentation

It is shown that, under the proposed RW loss framework, certain loss functions, such as Active Contour and Boundary loss, can be reformulated similarly with appropriate RW maps, thus revealing their underlying similarities and a new perspective to understand these loss functions.

Determining and Measuring the Amount of Region Having COVID-19 on Lung Images

The results obtained show that such an artificial intelligence-based method during the pandemic period will help prioritize and automate the diagnosis of COVID-19 patients.

Comparative study of deep learning methods for the automatic segmentation of lung, lesion and lesion type in CT scans of COVID-19 patients

This study compares twelve deep learning algorithms using a multi-center dataset, including both open-source and in-house developed algorithms, and shows that all methods perform binary lesion segmentation with an average volume error that is better than visual assessment by human raters, suggesting these methods are mature enough for a large-scale evaluation for use in clinical practice.

On Image Segmentation With Noisy Labels: Characterization and Volume Properties of the Optimal Solutions to Accuracy and Dice

We study two of the most popular performance metrics in medical image segmentation, Accuracy and Dice, when the target labels are noisy. For both metrics, several statements related to

Impact of Lung Segmentation on the Diagnosis and Explanation of COVID-19 in Chest X-ray Images

Experiments support the conclusion that even after segmentation in COVID-19 identification using CXR images, there is a strong bias introduced by underlying factors from different sources.

Loss odyssey in medical image segmentation

...

References

SHOWING 1-10 OF 24 REFERENCES

Generalised Dice overlap as a deep learning loss function for highly unbalanced segmentations

This work investigates 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 and proposes to use the class re-balancing properties of the Generalized Dice overlap as a robust and accurate deep-learning loss function for unbalanced tasks.

Morphometric analysis of white matter lesions in MR images: method and validation

The results indicate that, compared to a manual method, the use of the semiautomatic technique not only facilitates the analysis of the images, but also has similar or lower intra- and interrater variabilities.

The Lovasz-Softmax Loss: A Tractable Surrogate for the Optimization of the Intersection-Over-Union Measure in Neural Networks

This work presents a method for direct optimization of the mean intersection-over-union loss in neural networks, in the context of semantic image segmentation, based on the convex Lovász extension of submodular losses.

U-Net: Convolutional Networks for Biomedical Image Segmentation

It is shown that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks.

Optimizing Intersection-Over-Union in Deep Neural Networks for Image Segmentation

This paper proposes an approach for directly optimizing this intersection-over-union (IoU) measure in deep neural networks and demonstrates that this approach outperforms DNNs trained with standard softmax loss.

Adam: A Method for Stochastic Optimization

This work introduces Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments, and provides a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework.

DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

This work addresses the task of semantic image segmentation with Deep Learning and proposes atrous spatial pyramid pooling (ASPP), which is proposed to robustly segment objects at multiple scales, and improves the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models.

Artificial Intelligence for Medical Image Analysis: A Guide for Authors and Reviewers.

Artificial intelligence is in the early phases of application to medical imaging, and patient safety demands a commitment to sound methods and avoidance of rhetorical and overly optimistic claims.

Revisiting uncertainty in graph cut solutions

It is demonstrated empirically that - after proper training - uncertainties based on min-marginals provide better-calibrated probabilities than baselines and that these distributions can be exploited in a decision-theoretic way for improved segmentation in low-level vision.