-
Differentiable Rendering for 3D Fluorescence Microscopy
Authors:
Sacha Ichbiah,
Fabrice Delbary,
Hervé Turlier
Abstract:
Differentiable rendering is a growing field that is at the heart of many recent advances in solving inverse graphics problems, such as the reconstruction of 3D scenes from 2D images. By making the rendering process differentiable, one can compute gradients of the output image with respect to the different scene parameters efficiently using automatic differentiation. Interested in the potential of…
▽ More
Differentiable rendering is a growing field that is at the heart of many recent advances in solving inverse graphics problems, such as the reconstruction of 3D scenes from 2D images. By making the rendering process differentiable, one can compute gradients of the output image with respect to the different scene parameters efficiently using automatic differentiation. Interested in the potential of such methods for the analysis of fluorescence microscopy images, we introduce deltaMic, a microscopy renderer that can generate a 3D fluorescence microscopy image from a 3D scene in a fully differentiable manner. By convolving the meshes in the scene with the point spread function (PSF) of the microscope, that characterizes the response of its imaging system to a point source, we emulate the 3D image creation process of fluorescence microscopy. This is achieved by computing the Fourier transform (FT) of the mesh and performing the convolution in the Fourier domain. Naive implementation of such mesh FT is however slow, inefficient, and sensitive to numerical precision. We solve these difficulties by providing a memory and computationally efficient fully differentiable GPU implementation of the 3D mesh FT. We demonstrate the potential of our method by reconstructing complex shapes from artificial microscopy images. Eventually, we apply our renderer to real confocal fluorescence microscopy images of embryos to accurately reconstruct the multicellular shapes of these cell aggregates.
△ Less
Submitted 18 March, 2023;
originally announced March 2023.
-
Physical configurations of a cell doublet with line tension, a theoretical study
Authors:
Fabrice Delbary
Abstract:
As a first approximation, early embryos may be modeled as foams whose shape depends on the surface tensions of each cell. However it has been remarked that exist line tensions at polarized exterior cellular interfaces (apical). In order to understand the changes it may imply on the usual foam model, a simple case study is considered: a double cell with line tension. Phase diagrams, bifurcations, p…
▽ More
As a first approximation, early embryos may be modeled as foams whose shape depends on the surface tensions of each cell. However it has been remarked that exist line tensions at polarized exterior cellular interfaces (apical). In order to understand the changes it may imply on the usual foam model, a simple case study is considered: a double cell with line tension. Phase diagrams, bifurcations, possible new configurations are studied.
△ Less
Submitted 2 July, 2024; v1 submitted 7 September, 2022;
originally announced September 2022.
-
A generalized Newton iteration for computing the solution of the inverse Henderson problem
Authors:
Fabrice Delbary,
Martin Hanke,
Dmitry Ivanizki
Abstract:
We develop a generalized Newton scheme IHNC for the construction of effective pair potentials for systems of interacting point-like particles.The construction is made in such a way that the distribution of the particles matches a given radial distribution function. The IHNC iteration uses the hypernetted-chain integral equation for an approximate evaluation of the inverse of the Jacobian of the fo…
▽ More
We develop a generalized Newton scheme IHNC for the construction of effective pair potentials for systems of interacting point-like particles.The construction is made in such a way that the distribution of the particles matches a given radial distribution function. The IHNC iteration uses the hypernetted-chain integral equation for an approximate evaluation of the inverse of the Jacobian of the forward operator.
In contrast to the full Newton method realized in the Inverse Monte Carlo (IMC) scheme, the IHNC algorithm requires only a single molecular dynamics computation of the radial distribution function per iteration step, and no further expensive cross-correlations. Numerical experiments are shown to demonstrate that the method is as efficient as the IMC scheme, and that it easily allows to incorporate thermodynamical constraints.
△ Less
Submitted 28 June, 2018;
originally announced June 2018.
-
The Radon transform and the Hough transform: a unifying perspective
Authors:
Riccardo Aramini,
Fabrice Delbary,
Mauro C. Beltrametti,
Michele Piana,
Anna Maria Massone
Abstract:
The Radon transform is a linear integral transform that mimics the data formation process in medical imaging modalities like X-ray Computerized Tomography and Positron Emission Tomography. The Hough transform is a pattern recognition technique, which is mainly used to detect straight lines in digital images and which has been recently extended to the automatic recognition of algebraic plane curves…
▽ More
The Radon transform is a linear integral transform that mimics the data formation process in medical imaging modalities like X-ray Computerized Tomography and Positron Emission Tomography. The Hough transform is a pattern recognition technique, which is mainly used to detect straight lines in digital images and which has been recently extended to the automatic recognition of algebraic plane curves. Although defined in very different ways, in numerical applications both transforms ultimately take an image as an input and provide, as an output, a function defined on a parameter space. The parameters in this space describe a family of curves, which represent either the integration domains considered in the (generalized) Radon transform, or the curves to be detected by means of the Hough transform. In both cases, the 2D plot of the intensity values of the output function is the so-called (Radon or Hough) sinogram. While the Hough sinogram is produced by an algorithm whose implementation requires that the parameter space be discretized in cells, the Radon sinogram is mathematically defined on a continuous parameter space, which in turn may need to be discretized just for physical or numerical reasons. In this paper, by considering a more general and n-dimensional setting, we prove that, whether the input image is described as a set of points (possibly with different intensity values) or as a piecewise constant function, its (rescaled) Hough sinogram converges to the corresponding Radon sinogram as the discretization step in the parameter space tends to zero. We also show that this result may have a notable impact on the image reconstruction problem of inverting the Radon sinogram recorded by a medical imaging scanner, and that the description of the Hough transform problem within the framework of regularization theory for inverse problems is worth investigating.
△ Less
Submitted 30 May, 2016;
originally announced May 2016.
-
Compartmental analysis of dynamic nuclear medicine data: models and identifiability
Authors:
Fabrice Delbary,
Sara Garbarino,
Valentina Vivaldi
Abstract:
Compartmental models based on tracer mass balance are extensively used in clinical and pre-clinical nuclear medicine in order to obtain quantitative information on tracer metabolism in the biological tissue. This paper is the first of a series of two that deal with the problem of tracer coefficient estimation via compartmental modelling in an inverse problem framework. Specifically, here we discus…
▽ More
Compartmental models based on tracer mass balance are extensively used in clinical and pre-clinical nuclear medicine in order to obtain quantitative information on tracer metabolism in the biological tissue. This paper is the first of a series of two that deal with the problem of tracer coefficient estimation via compartmental modelling in an inverse problem framework. Specifically, here we discuss the identifiability problem for a general n-dimension compartmental system and provide uniqueness results in the case of two-compartment and three-compartment compartmental models. The second paper will utilize this framework in order to show how non-linear regularization schemes can be applied to obtain numerical estimates of the tracer coefficients in the case of nuclear medicine data corresponding to brain, liver and kidney physiology.
△ Less
Submitted 2 December, 2016; v1 submitted 7 March, 2016;
originally announced March 2016.
