Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Registration Uncertainty to Segmentation Uncertainty

Junyu Chen, Yihao Liu, Shuwen Wei, Zhangxing Bian, Aaron Carass, Yong Du

TL;DR

This work tackles the mismatch between registration uncertainty and segmentation error in label propagation by introducing two complementary uncertainty estimates: an epistemic segmentation uncertainty defined as the entropy of the propagated labels and an aleatoric segmentation uncertainty produced by a compact DNN conditioned on appearance differences between warped and fixed images. The latter uses a $L_{\beta-NLL}$ objective with $\beta=1$ to estimate label-propagation variance $\sigma^2$ without requiring labels at test time. Experiments on 3D cardiac MRI datasets show that both segmentation uncertainties correlate with label-propagation errors and yield superior Dice scores compared with baselines. By bridging registration and segmentation uncertainty, the approach provides enhanced insight into where segmentation errors may occur and has potential applications in atlas-based segmentation and dosimetric uncertainty in cancer therapy.

Abstract

Understanding the uncertainty inherent in deep learning-based image registration models has been an ongoing area of research. Existing methods have been developed to quantify both transformation and appearance uncertainties related to the registration process, elucidating areas where the model may exhibit ambiguity regarding the generated deformation. However, our study reveals that neither uncertainty effectively estimates the potential errors when the registration model is used for label propagation. Here, we propose a novel framework to concurrently estimate both the epistemic and aleatoric segmentation uncertainties for image registration. To this end, we implement a compact deep neural network (DNN) designed to transform the appearance discrepancy in the warping into aleatoric segmentation uncertainty by minimizing a negative log-likelihood loss function. Furthermore, we present epistemic segmentation uncertainty within the label propagation process as the entropy of the propagated labels. By introducing segmentation uncertainty along with existing methods for estimating registration uncertainty, we offer vital insights into the potential uncertainties at different stages of image registration. We validated our proposed framework using publicly available datasets, and the results prove that the segmentation uncertainties estimated with the proposed method correlate well with errors in label propagation, all while achieving superior registration performance.

From Registration Uncertainty to Segmentation Uncertainty

TL;DR

This work tackles the mismatch between registration uncertainty and segmentation error in label propagation by introducing two complementary uncertainty estimates: an epistemic segmentation uncertainty defined as the entropy of the propagated labels and an aleatoric segmentation uncertainty produced by a compact DNN conditioned on appearance differences between warped and fixed images. The latter uses a objective with to estimate label-propagation variance without requiring labels at test time. Experiments on 3D cardiac MRI datasets show that both segmentation uncertainties correlate with label-propagation errors and yield superior Dice scores compared with baselines. By bridging registration and segmentation uncertainty, the approach provides enhanced insight into where segmentation errors may occur and has potential applications in atlas-based segmentation and dosimetric uncertainty in cancer therapy.

Abstract

Understanding the uncertainty inherent in deep learning-based image registration models has been an ongoing area of research. Existing methods have been developed to quantify both transformation and appearance uncertainties related to the registration process, elucidating areas where the model may exhibit ambiguity regarding the generated deformation. However, our study reveals that neither uncertainty effectively estimates the potential errors when the registration model is used for label propagation. Here, we propose a novel framework to concurrently estimate both the epistemic and aleatoric segmentation uncertainties for image registration. To this end, we implement a compact deep neural network (DNN) designed to transform the appearance discrepancy in the warping into aleatoric segmentation uncertainty by minimizing a negative log-likelihood loss function. Furthermore, we present epistemic segmentation uncertainty within the label propagation process as the entropy of the propagated labels. By introducing segmentation uncertainty along with existing methods for estimating registration uncertainty, we offer vital insights into the potential uncertainties at different stages of image registration. We validated our proposed framework using publicly available datasets, and the results prove that the segmentation uncertainties estimated with the proposed method correlate well with errors in label propagation, all while achieving superior registration performance.
Paper Structure (8 sections, 3 equations, 2 figures, 1 table)

This paper contains 8 sections, 3 equations, 2 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. METHODS
  3. Epistemic Segmentation Uncertainty
  4. Aleatoric Segmentation Uncertainty
  5. Experiments
  6. Results
  7. Conclusions
  8. Acknowledgments

Figures (2)

  • Figure 1: The overall framework of estimating registration and segmentation uncertainty for DNN-based image registration.
  • Figure 2: Qualitative results of various registration methods, as well as different schemes for registration and segmentation uncertainty quantification using the proposed method. The upper panel shows the qualitative results from different registration methods. The upper panel presents qualitative comparisons across different registration methods. The bottom panel illustrates various uncertainty quantification metrics: the first image depicts the absolute difference in aligned images; yellow highlights registration uncertainties related to transformation and appearance; red indicates label propagation error as squared errors per class; green represents epistemic segmentation uncertainty; and blue delineates aleatoric segmentation uncertainty.