Doctor's Theses (authored and supervised):
"Analysis of 3D and 4D Images of Organisms in Embryogenesis";
Supervisor, Reviewer: M. Sramek, P. Frolkovic;
Institut für Computergraphik und Algorithmen,
oral examination: 2015-08-24.
In this work, we present a few modifications to the state-of-the-art algorithms, as well as several novel approaches, related to the detection of cells in biological image processing.
We start by explanation of a PDE-based image processing evolution called FBLSCD and study its properties. We then define a fully automatic way of finding the stop time for this evolution. Afterwards, we try to see the FBLSCD as a morphological grayscale erosion, and we formulate a novel cell detection algorithm, called LSOpen, as an intersection of PDE-based and morphological image processing schools.
Then, we discuss the best ways of inspecting cell detection results, i.e. cell identifiers. We try to quantitatively benchmark various cell detection methods by the relative amount of false positives, false negatives and multiply-detected centers yielded. We will observe that comparing cell detection results in a binary fashion is insufficient, therefore we are going to utilize the concept of distance function.
Motivated by this need for robust cell detection result comparison, we analyze commonly-used methods for computing the distance function and afterwards we formulate a novel algorithm. This one has complexity O(n log2 n) and it yields Euclidean distance. In addition to that, we introduce a modification to this algorithm, enabling it to work also in maze-like, wall- and corner-containing, environments.
This modification relies on the line rasterization algorithm. We perform various experiments to study and compare distance function methods. Results illustrate the viability of newly-proposed method.
Further, a software for the comparing and inspecting cell detection results, SliceViewer, is specified, designed, implemented and tested.
In the end, quantitative experiments are discussed, validating the above-mentioned novelties.
Electronic version of the publication:
Created from the Publication Database of the Vienna University of Technology.