Robust identification of topological defects in discrete vector fields with applications to biological image data.

Authors Karl Hoffmann
University Technische Universit├Ąt Dresden
Examination Date 2022-12-19
Open Access false
Print Publication Date 2022-12-19
Online Publication Date 2022-12-19
Abstract Topological defects are distinguished objects in vector fields that occur in a wide range of applications, ranging from material sciences to cosmology to bio-medical imaging and fingerprint recognition. This thesis considers topological point defects, also known as singular points, of two-dimensional vector fields. Besides Euclidean vectors as representation of modulus and direction, this also includes nematic vectors that equally have a modulus but direction is replaced with a head-to-tail symmetric orientation. In both case, a singular point or topological defect is an isolated discontinuity in an otherwise continuous vector field. It is characterized by its index or topological charge, which attains integer values for polar and half-integer values for nematic vector fields. There are different yet equivalent approaches to define the index. They either base on homology groups and the Brouwer degree, or on the first fundamental group and the mapping degree, or relatedly on lifting of a loop path enclosing the singular point. The definition by lift used here translates changes in the vector field along a path into a summed change in orientation angle. This translates to topological defects in discretized vector fields, where topological charge is calculated as sum of finite angle differences along a loop path between discretization points. On closer inspection, this calculation is an estimation, and is guaranteed to yield the correct estimate only with additional assumptions, for example when the underlying continuous-domain vector field is smooth and sampled at sufficiently high spatial resolution. Otherwise, arbitrary locations and charges of topological defects are possible, which yield exactly the same discretized vectors by the periodicity of representative orientation angles. Besides, the estimated topological charge depends discontinuously on each of the discrete input vectors and exhibits discrete jumps. As application data typically is subject to noise and uncertainty, this raises the question how reliable are topological defects identified in it. The present thesis quantifies, how large perturbations of a vector field are admissible without alteration of topological defects and charges. To that end, it introduces a robustness measure for each edge in a discretization grid that are combined along loop paths. Replacing critical edges of minimal robustness within a loop path by other path segments around a minimally larger area allows targeted increase of robustness. This data-dependent method called expansion over the critical edge is iterated until a user-set robustness is satisfied. The final areas of this algorithm are shown to have minimal size and therefore maximal spatial resolution, which also adapts to the local quality of data. The areas are also given as the faces in the graph of sufficiently robust edges after deleting all vertices of degree 1 (leaves) and all their connected edges. The minimal robust areas turn out to be nested by inclusion according to their robustness threshold. This allows to tradeoff detection robustness of topological charges versus their localization accuracy, both within a selection of pre-defined loop path shapes, and for free data-dependent expansion over the critical edge. Differently from defect identification by pattern matching, there is no restriction on the charge detectable. Besides, the robustness is shown to detect the size of unordered cores of defects. Robust defect areas indicate possible defect dynamics comprising motion, defect pair generation and annihilation already from single time point data. The robustness is also applicable to irregular discretization grids thanks to its graph theoretic characterization, and an extension to curved surfaces is foreseeable. The robust data-dependent defect identification is exemplified on microscopy images of the fruit fly Drosophila melanogaster. During Dorsal Closure, a developmental process, a cell sheet called amnioserosa contracts in highly regulated manner, whereby forces are actively generated and propagated along filamentous proteins like actin. Thereby, activity level and visco-elastic properties of the tissue are linked to the topological defects in the actin orientation field. Robust detection of these reveals that the sum over robust charges is clearly positive in the hundreds, whereas the overall sum of charge without robustness consideration fluctuates around zero. Numerous charges are observed, but $\pm 1/2$ dominate and confirm the amnioserosa as nematic material despite polar molecular constituents like actin. The sizes of robust defects span three orders of magnitude, and the largest defects follow the shapes of biological cells. The size distribution decays by a power law with the power for positive defects being more negative. Time courses show slightly higher speed of motion for +1/2 defects than for -1/2 defects, an order of magnitude above material flow velocity. Experiments with a genetic modification in the protein Crumbs had shown excess contraction of the amnioserosa cell layer during development. Comparing defect velocity of these embryos to wildtype suggests that viscosity and rotational viscosity increase stronger than activity level. This hypothesis remains to be tested in a combination of experiments and simulations, yet it would not have been generated in the first place without consideration of robust defects. More generally, the presented robustness measure and optimal data-dependent identification of topological defects could benefit the analysis of defects in discretized vector fields in a variety of disciplines. The optimal data-dependent identification allows for example to calculate error distributions for charge and localization of defects. The size, shape, and nested inclusion of robust defects constitute new observables, that generate numerous follow-up questions already for the fruit fly and enable novel analyses.
Cover Image
Affiliated With Sbalzarini
Selected By
Acknowledged Services
Publication Status Published
Edoc Link
Sfx Link
PubMed ID
WebOfScience Link
Alternative Full Text URL
Display Publisher Download Only false
Visible On MPI-CBG Website true
PDF Downloadable true
Created By thuem
Added Date 2023-06-12
Last Edited By thuem
Last Edited Date 2023-06-12 16:10:36.588
Library ID 8566
Document ID PB 527
Entry Complete true
eDoc Compliant true
Include in Edoc Report true
In Pure false
Ready for eDoc Export false
Author Affiliations Complete false
Project Name
Project URL
Grant ID
Funding Programme
Funding Organisation