Contactless estimation of continuum displacement and mechanical compressibility from image series using a deep learning based framework

Anto Nivin Maria Antony, Thomas Richter, Evgeny Gladilin · 2026

Read Contactless estimation of continuum displacement and mechanical compressibility from image series using a deep learning based framework by Anto Nivin Maria Antony, Thomas Richter, Evgeny Gladilin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07065.txt.

Introduction

A. N. Maria Antony and two co-authors estimate material traits from image series alone, with no touch on the sample. Contactless measurement matters in engineering and medicine where direct probing is impossible. Old approaches rely on slow iterative alignment algorithms and numerical solvers, such as finite element methods. Those do not suit high-throughput data.

Their deep learning framework uses two nets. One handles image alignment. One estimates how much a sample can be squeezed. Together they yield shifts of the continuum and squeeze traits end to end from the image series. The framework beat old approaches on both speed and score. Notably, it gauged squeeze traits well even when the alignment map strayed far from the reference shift field. The authors credit higher-order traits, such as swirl in the vector field, rather than plain local traits of shift.

Catalog record

CREATORAnto Nivin Maria Antony, Thomas Richter, Evgeny Gladilin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07065

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Contactless estimation of continuum displacement and mechanical compressibility from image series using a deep learning based framework Anto Nivin Maria Antony, Thomas Richter, Evgeny Gladilin Abstract Contactless and non-invasive estimation of mechanical properties of physical media from optical observations is of interest for manifold engineering and biomedical applications, where direct physical measurements are not possible. Conventional approaches to the assessment of image displacement and non-contact material probing typically rely on time-consuming iterative algorithms for non-rigid image registration and constitutive modelling using discretization and iterative numerical solving techniques, such as Finite Element Method (FEM) and Finite Difference Method (FDM), which are not suitable for high-throughput data processing. Here, we present an efficient deep learning based end-to-end approach for the estimation of continuum displacement and material compressibility directly from the image series. Based on two deep neural networks for image registration and material compressibility estimation, this framework outperforms conventional approaches in terms of efficiency and accuracy. In particular, our experimental results show that the deep learning model trained on a set of reference data can accurately determine the material compressibility even in the presence of substantial local deviations of the mapping predicted by image registration from the reference displacement field. Our findings suggest that the remarkable accuracy of the deep learning end-to-end model originates from its ability to assess higher-order cognitive features, such as the vorticity of the vector field, rather than conventional local features of the image displacement. 1 Introduction Contactless analysis of mechanical structures, such as engineering materials and biological tissues, is a method of choice when direct physical probing is not possible or applicable [ 29 , 22 ] . This is, in particular, the case when target structures can only be monitored remotely using non-invasive optical observation techniques such as magnetic resonance elastography [ 23 ] , ultrasound elastography [ 24 ] , ultrasound shear wave elastography [ 28 ] , etc. Non-contacting optical monitoring enables significantly higher throughput of measurements compared to contacting techniques; however, it provides indirect observations in the form of image sequences and requires additional image analysis and numerical simulation to reveal hidden parameters of underlying constitutive models. In a number of previous works, including the author’s early findings [ 9 , 10 , 8 ] , approaches to image- and model-based estimation of material properties of optically monitored structures was reported. These approaches are typically based on image registration followed by fitting a numerical model of material behaviour to a pair of co-registered images. Depending on the particular data modality and task, whole image or boundary (contour re. surface) based non-rigid registration in combination with conventional iterative solving techniques such as Finite Difference or Finite Element Method is applied. The pipelines of conventional image analysis and numerical problem solving are, however, known to be time-consuming and cumbersome (especially with respect to individual mesh generation), which substantially limits their high-throughput capabilities [ 36 , 27 , 5 ] . In contrast, modern screening methods generate a large amount of real-time images that require efficient high-throughput algorithmic solutions. The understanding of the material properties of cells is the fundamental basis in the fields of cell mechanics and mechanobiology [ 17 ] . The biological behaviours of the cells, like division, deformation, and differentiation, are related to the material properties of cells [ 20 , 12 ] . The material property is mostly explained with the help of Young’s modulus and Poisson’s ratio [ 17 , 14 , 15 ] . The stiffness is one parameter that is extensively studied, given by Young’s modulus [ 21 ] . However, the Poisson’s ratio which explains the material compressibility is mostly overlooked [ 17 ] . The estimation of Poisson’s ratio is crucial for cancer diagnosis, prognosis, and treatment as explained in [ 14 , 15 ] . In the past decades, numerous image registration methods for uni- and multi-model rigid and non-rigid alignment were developed [ 37 , 34 , 6 , 33 ] . Conventional methods of image registration are computationally expensive [ 18 ] , in particular, when applied to images of high resolution when convergence of the iterative optimization solvers becomes slow. Recent advances in machine learning, in particular, deep learning, open new opportunities for efficient computational solutions. Deep learning has been used for efficient implementation of 2D and 3D image registration [ 4 , 1 ] . However, investigations of the accuracy of deep learning registration methods in application to concrete registration problems are rarely performed. The application of deep learning to continuum and computational mechanics has been comprehensively reviewed in [ 2 , 11 ] . In [ 35 ] , the stress-strain patterns were studied with Convolutional Neural Networks (CNN) along with principal component analysis, and in [ 25 ] it is studied in digital composites using Fourier Neural Operator (FNO) in an end to end fashion. The studies were performed in a supervised fashion with a pair of input geometries and ground truth solutions. To our best knowledge, determining Poisson’s ratio from images hasn’t been studied yet. In this study, we present a framework for fully automated estimation of material compressibility of optically monitored structures which is based on a combination of deep learning non-rigid image registration trained on a pair of source target-images and deep learned-based regression model ( [ 26 ] ) trained to assign 2D whole displacement fields generated from Finite Difference scheme to one of the three material compressibilities corresponding to low, high and intermediate compressibility. Throughout the study, we assume the material medium is isotropic and homogeneous. The manuscript is organized as follows: first, we describe the basic methods of this study, including the linear elastic Finite Difference Method, deep learning based non-rigid image registration scheme, as well as the deep learning-based model for material compressibility estimation. Then, the results of our investigations are based on the reconstruction of the displacement field from images and, the estimation of material compressibility followed by discussion. 2 Methods This section presents the methods we used for generation of reference data (i.e. linear elastic displacement fields, deformed images), deep learning based non-rigid image registration, and estimation of Poisson’s ratio. 2.1 Linear elastic deformation model - 2D Lamé-Navier PDE In the case of a one-material problem and boundary forces that are given implicitly as prescribed boundary displacements (i.e. \textbf{f}=0 and \textbf{u}=\textbf{u}_{\text{known}}+\textbf{u}_{\text{unknown}} ), Eq. 1 takes a quasi-geometric form \,-(1-2\nu)\,\boldsymbol{\Delta}{u}-\text{grad}\,\text{div}\,{u}=0\,, (1) with the displacement {\bf u} as a single variable and the Poisson’s ratio \nu as single material parameter. Eq. 1 with arbitrary boundary conditions is solved using the Finite Difference Method on the regular 2D image grid. In the case of a quasi-geometrical (pure displacement), the Finite Difference scheme of the Lamé-Navier PDE takes the following form \begin{gathered}2(1-\nu)\left[\frac{u_{x}(i+1,j)-2u_{x}(i,j)+u_{x}(i-1,j)}{\Delta x^{2}}\right]\ +\ (1-2\nu)\left[\frac{u_{x}(i,j+1)-2u_{x}(i,j)+u_{x}(i,j-1)}{\Delta y^{2}}\right]\\ +\ \left[\frac{u_{y}(i+1,j+1)-u_{y}(i+1,j-1)-u_{y}(i-1,j+1)+u_{y}(i-1,j-1)}{4\Delta x\Delta y}\right]\ =0\,,\\ 2(1-\nu)\left[\frac{u_{y}(i,j+1)-2u_{y}(i,j)+u_{y}(i,j-1)}{\Delta y^{2}}\right]\ +\ (1-2\nu)\left[\frac{u_{y}(i+1,j)-2u_{y}(i,j)+u_{y}(i-1,j)}{\Delta x^{2}}\right]\\ +\ \left[\frac{u_{x}(i+1,j+1)-u_{x}(i+1,j-1)-u_{x}(i-1,j+1)+u_{x}(i-1,j-1)}{4\Delta x\Delta y}\right]\ =0\,.\end{gathered} (2) The FDM scheme transforms the continuous second-order 2D Lamé-Navier PDE to the linear system of equations with respect to unknown values of xy-components of the displacement: A\,{u}={b}\,, (3) where A is a sparse symmetric and positive definite matrix (the so-called stiffness matrix), {b} is the right-hand side vector resulting from the implementation of known boundary conditions, and {u} is the vector of

Read the complete text