Feasibility and quantitative analysis of a biomechanical model-guided lung elastography for radiotherapy

Abstract

We present a novel methodology for performing lung elastography within the radiotherapy context. Our approach employed a physics-based model and a novel convergence magnification approach to estimate the lung elasticity distribution. For a systematic analysis, we employed a physics-based virtual lung phantom with CT source geometry and a heterogeneous voxel-to-voxel elasticity distribution. A set of 18 synthetic CT image datasets with known ground-truth elasticity representing normal, emphysematous, and tumor tissue within the lungs was generated and used as input for the lung elastography. During the lung elastography, we start by re-estimating the lung deformations using a deformable image registration procedure. For known phantom geometry, boundary conditions and lung deformation, we solve for the elasticity distribution by iteratively optimizing the tissue elasticity and deforming the lung model for given boundary constraints. To improve the estimation accuracy, a convergence magnification approach was formulated using a physics-based process that amplified the geometrical differences between the ground-truth and the deformed geometry. Our results showed that the model-guided approach estimated the elasticity with 91.94±5.20% of voxels within 0.5 mm of ground-truth displacement and 1 kPa of the ground-truth elasticity. The novel convergence technique presented in this paper improved the accuracy of the estimated elasticity from 77.05% to 91.94%. In the systematic analysis, variations in forward model, ground-truth elasticity distribution, boundary deformation, and geometry were investigated with respect to their effect on the accuracy of the elasticity estimation technique.

An adversarial machine learning framework and biomechanical model‐guided approach for computing 3D lung tissue elasticity from end‐expiration 3DCT

“…For the validation dataset, consisting of 13 4D datasets, we were able to obtain an accuracy of 0.87 ± 0.4 KPa as tabulated in Table II. Our previous studies for elastography showed that a difference of <1 KPa interval yields clinically agreeable deformation accuracy 20,33 . Since elasticity is known to have a range of 0–12 KPa as documented in two of our previous papers, the current error is <8% of the expected tissue elasticity range.…”

“…The correctness and importance of performing an inverse elasticity distribution has been investigated and discussed in Refs. [12,20,34]. Specifically, the estimated elasticity was shown as a potential image‐based biomarker for characterizing the degree of chronic obstructive pulmonary disease (COPD) 34 .…”

“…As previously discussed in Refs. [19,20] no specific external forces were used for this purpose. The biomechanical lung model was actuated by using the lung surface displacements as pointwise and fixed boundary constraints.…”

Estimation and validation of patient‐specific high‐resolution lung elasticity derived from 4DCT

“…The model has been previously described in Ref. and specifically applied to the lungs . This approach is appropriate for our current study as its applicability has been systematically investigated .…”

“…[26,27] and specifically applied to the lungs. 28,29 This approach is appropriate for our current study as its applicability has been systematically investigated. 29 For clarity purposes, a brief description of the constitutive equations for the forward model 27 is presented here.…”