Lymphoma segmentation from 3D PET-CT images using a deep evidential network

Ling, Hoe I.; Ruan, Su; Decazes, Pierre; Denœux, Thierry

doi:10.1016/j.ijar.2022.06.007

Cited by 30 publications

(27 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…. , (x n , y n )}, the accuracy of the predictions can be measured by the empirical risk (11) introduced in Section III-A. To train the ENNreg model described in Section IV-A, we will use the regularized average loss…”

Section: B Learningmentioning

confidence: 99%

See 1 more Smart Citation

Quantifying Prediction Uncertainty in Regression using Random Fuzzy Sets: the ENNreg model

Denœux¹

2023

Preprint

View full text Add to dashboard Cite

<p>We introduce a neural network model for regression in which prediction uncertainty is quantified by Gaussian random fuzzy numbers (GRFNs), a newly introduced family of random fuzzy subsets of the real line that generalizes both Gaussian random variables and Gaussian possibility distributions. The output GRFN is constructed by combining GRFNs induced by prototypes using a combination operator that generalizes Dempster's rule of Evidence Theory. The three output units indicate the most plausible value of the response variable, variability around this value, and epistemic uncertainty. The network is trained by minimizing a loss function that generalizes the negative log-likelihood. Comparative experiments show that this method is competitive, both in terms of prediction accuracy and calibration error, with state-of-the-art techniques such as random forests or deep learning with Monte Carlo dropout. In addition, the model outputs a predictive belief function that can be shown to be calibrated, in the sense that it allows us to compute conservative prediction intervals with specified belief degree.</p>

show abstract

Section: B Learningmentioning

confidence: 99%

“…(ENN) model introduced in [9], in which the learning vectors are summarized by prototypes, whose location in feature space is optimized together with other network parameters. Recently, this idea has been applied to deep networks [10] [11] by adding a "DS layer" to a deep architecture.…”

Section: Introductionmentioning

confidence: 99%

Quantifying Prediction Uncertainty in Regression using Random Fuzzy Sets: the ENNreg model

Denœux¹

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Existing uncertainty estimation methods remain poorly utilized in medical image segmentation. Uncertainty quantification methods in medical domain include Bayesian- [20], ensemble- [21], evidential- [22], and deterministic-based methods [23,24]. A simple way to produce uncertainty for medical image segmentation is to use an ensemble of deep networks [25,26].…”

Section: Introductionmentioning

confidence: 99%

“…Unfortunately, the above methods inevitably change the network structure and incur computational costs. A recent study has proposed using a deep feature-extraction module and an evidential layer to segment lymphomas from positron emission tomography and computed tomography image [22]. The main aims of these studies remained on guiding uncertainty to improve segmentation performance rather than obtaining more robust segmentation with calibrated uncertainty, and on generating uncertainty to evaluate the segmentation results rather than utilizing the calibrated uncertainty to further optimize the model training.…”

Section: Introductionmentioning

confidence: 99%

EvidenceCap: Towards trustworthy medical image segmentation via evidential identity cap

Yuan

Shen

Chen

et al. 2023

Preprint

View full text Add to dashboard Cite

Medical image segmentation (MIS) is essential for supporting disease diagnosis and treatment effect assessment. Despite considerable advances in artificial intelligence (AI) for MIS, clinicians remain skeptical of its utility, maintaining low confidence in such black box systems, with this problem being exacerbated by low generalization for out-of-distribution (OOD) data. To move towards effective clinical utilization, we propose a foundation model named EvidenceCap, which makes the box transparent in a quantifiable way by uncertainty estimation. EvidenceCap not only makes AI visible in regions of uncertainty and OOD data, but also enhances the reliability, robustness, and computational efficiency of MIS. Uncertainty is modeled explicitly through subjective logic theory to gather strong evidence from features. We show the effectiveness of EvidenceCap in three segmentation datasets and apply it to the clinic. Our work sheds light on clinical safe applications and explainable AI, and can contribute towards trustworthiness in the medical domain.

show abstract

“…The quantification of prediction uncertainty in Machine Learning has recently gained a lot of attention (see, e.g., [1]- [4]). Whereas most approaches are based on Bayesian inference, other theories of uncertainty, such as Dempster-Shafer (DS) theory [5], [6], have also proved to be very promising [7], [8]. DS theory of belief functions, also known as theory of belief functions of evidence theory, is a mathematical formalism for reasoning with uncertainty [5], [6], which makes it possible to overcome some limitations of Bayesian inference.…”

Section: Introductionmentioning

confidence: 99%

Quantifying Prediction Uncertainty in Regression Using Random Fuzzy Sets: The ENNreg Model

Denœux

2023

IEEE Trans. Fuzzy Syst.

View full text Add to dashboard Cite

We introduce a neural network model for regression in which prediction uncertainty is quantified by Gaussian random fuzzy numbers (GRFNs), a newly introduced family of random fuzzy subsets of the real line that generalizes both Gaussian random variables and Gaussian possibility distributions. The output GRFN is constructed by combining GRFNs induced by prototypes using a combination operator that generalizes Dempster's rule of Evidence Theory. The three output units indicate the most plausible value of the response variable, variability around this value, and epistemic uncertainty. The network is trained by minimizing a loss function that generalizes the negative log-likelihood. Comparative experiments show that this method is competitive, both in terms of prediction accuracy and calibration error, with state-of-the-art techniques such as random forests or deep learning with Monte Carlo dropout. In addition, the model outputs a predictive belief function that can be shown to be calibrated, in the sense that it allows us to compute conservative prediction intervals with specified belief degree.

show abstract

Lymphoma segmentation from 3D PET-CT images using a deep evidential network

Cited by 30 publications

References 54 publications

Quantifying Prediction Uncertainty in Regression using Random Fuzzy Sets: the ENNreg model

Quantifying Prediction Uncertainty in Regression using Random Fuzzy Sets: the ENNreg model

EvidenceCap: Towards trustworthy medical image segmentation via evidential identity cap

Quantifying Prediction Uncertainty in Regression Using Random Fuzzy Sets: The ENNreg Model

Contact Info

Product

Resources

About