Analysis of strain estimation methods in phase-sensitive compression optical coherence elastography

Abstract

In compression optical coherence elastography (OCE), deformation is quantified as the local strain at each pixel in the OCT field-of-view. A range of strain estimation methods have been demonstrated, yet it is unclear which method provides the best performance. Here, we analyze the two most prevalent strain estimation methods used in phase-sensitive compression OCE, i.e., weighted least squares (WLS) and the vector method. We introduce a framework to compare strain imaging metrics, incorporating strain sensitivity, strain signal-to-noise ratio (SNR), strain resolution, and strain accuracy. In addition, we propose a new phase unwrapping algorithm in OCE, fast phase unwrapping (FPU), and combine it with WLS, termed WLSFPU. Using the framework, we compare this new strain estimation method with both a current implementation of WLS that incorporates weighted phase unwrapping (WPU), termed WLSWPU, and the vector method. Our analysis reveals that the three methods provide similar strain sensitivity, strain SNR, and strain resolution, but that WLSFPU extends the dynamic range of accurate, measurable local strain, e.g., measuring a strain of 2.5 mɛ with ∼4% error, that is ×11 and ×15 smaller than the error measured using WLSWPU and the vector method, respectively. We also demonstrate, for the first time, the capability to detect sub-resolution contrast in compression OCE, i.e., changes in strain occurring within the strain axial resolution, and how this contrast varies between the different strain estimation methods. Lastly, we compare the performance of the three strain estimation methods on mouse skeletal muscle and human breast tissue and demonstrate that WLSFPU avoids strain imaging artifacts resulting from phase unwrapping errors in WLSWPU and provides improved contrast over the other two methods.

Quantitative Compression Elastography With an Uncalibrated Stress Sensor

“…The algorithm used here will determine the key factors that must be considered. Like the estimation of speed, the window across which the strain is estimated will also influence the accuracy [40,41].…”

Tumor spheroid elasticity estimation using mechano-microscopy combined with a conditional generative adversarial network

“…Axial strain, 𝜀 𝑧 , is calculated by measuring the axial displacement, 𝑢 𝑧 , due to compression at every (𝑥, 𝑧) location and computing the gradient of 𝑢 𝑧 with respect to 𝑧, i.e., . Methods for computing the gradient include linear regression [27,28] and finite difference [50] methods; however, the presence of noise typically necessitates increasing the fitting range or performing spatial averaging to improve sensitivity, at the expense of resolution [38,39].…”

“…Methods for computing the gradient include linear regression [27,28] and finite difference [50] methods; however, the presence of noise typically necessitates increasing the fitting range or performing spatial averaging to improve sensitivity, at the expense of resolution [38,39].…”

“…where the phase difference returns to 𝜋 instead of decreasing past −𝜋 . Previously, phase unwrapping algorithms have been used to convert phase difference to displacement [27,28,32],…”

“…Another advantage of cGANs is that they can be trained to learn the direct correspondence between phase difference images and elasticity images. This avoids needing to perform phase unwrapping to produce displacement images in the algebraic and iterative methods, which can produce artefacts [28]. It also avoids the need for strain estimation, which is necessary in the algebraic method.…”