Quantifying the unknown impact of segmentation uncertainty on image-based simulations

Michael C Krygier,Lincoln N Collins,Scott A Roberts,Partha P Mukherjee,Krish Sharma,Carianne Martinez,Tyler Labonte,Chance Norris

doi:10.1038/s41467-021-25493-8

Abstract

Image-based simulation, the use of 3D images to calculate physical quantities, relies on image segmentation for geometry creation. However, this process introduces image segmentation uncertainty because different segmentation tools (both manual and machine-learning-based) will each produce a unique and valid segmentation. First, we demonstrate that these variations propagate into the physics simulations, compromising the resulting physics quantities. Second, we propose a general framework for rapidly quantifying segmentation uncertainty. Through the creation and sampling of segmentation uncertainty probability maps, we systematically and objectively create uncertainty distributions of the physics quantities. We show that physics quantity uncertainty distributions can follow a Normal distribution, but, in more complicated physics simulations, the resulting uncertainty distribution can be surprisingly nontrivial. We establish that bounding segmentation uncertainty can fail in these nontrivial situations. While our work does not eliminate segmentation uncertainty, it improves simulation credibility by making visible the previously unrecognized segmentation uncertainty plaguing image-based simulation.

Highlights

Image-based simulation, the use of 3D images to calculate physical quantities, relies on image segmentation for geometry creation
Image-based simulation is the process of performing quantitative numerical calculations, such as 3D finite-element simulations, on geometries constructed directly from 3D imaging techniques, including X-ray computed tomography (CT) and scanning electron microscopy
We begin by illustrating the EQUIPS workflow for quantifying segmentation uncertainty and propagating it to physics simulations on the exemplar of a woven composite material (Fig. 2)

Summary

Introduction

Image-based simulation, the use of 3D images to calculate physical quantities, relies on image segmentation for geometry creation This process introduces image segmentation uncertainty because different segmentation tools (both manual and machinelearning-based) will each produce a unique and valid segmentation. The outcome of this image-based simulation (Fig. 2c) is a single value for the physics quantity of interest (Fig. 2d) For this process (Fig. 2b), traditional image segmentation approaches involve manual segmentation, whereby a person applies a combination of image filtering techniques (e.g., smoothing, noise removal, contrast enhancement, or non-local means filters) and segmentation algorithms (e.g., simple thresholding, watershed, or multi-Otsu thresholding algorithms) to segment the image. CNNs have gained immense popularity for image segmentation in a variety of applications, including in energy storage[30], materials analyses[31,32,33], and medical diagnosis[34,35]

Methods

Results

Conclusion