Recent statistical analysis has suggested that accurate measurements of certain features of the placental chorionic surface vascular network (PCSVN) after birth may be useful in identifying risks of certain neurodevelopmental disorders, including Autism Spectrum Disorder. These features, taken from a single top-down photograph, can be extracted from a manual tracing of the PCSVN, but the tracing procedure itself is very time and resource intensive. Automating this tracing procedure would not only allow more data acquisition to study potential links between placental development and neonatal health, but even perhaps provide a real-time diagnostic for certain risk factors.
Much work has been to develop reliable vascular network segmentation methods for well-known image domains (such as retinal MRA images) using Hessian-based filters, namely the multiscale Frangi filter. It is desirable to extend these techniques to our image domain, but this approach has been historically hindered by the comparative irregularity of the placental surface as a whole, which introduces significant noise into filtered result. Here we provide an in-depth mathematical background of the multiscale Frangi filter, touching upon topics in differential geometry, linear scale space theory, multiscale methods, and morphological image processing. Informed by this theory, we are able to identify stricter parameters that allow us to greatly improve our result. We also reimplement the Frangi filter in frequency space (using a Fast Fourier Transform), which allows a major speedup compared to previous efforts.
Compared to recent success in solving this problem using more modern techniques such as neural networks, we demonstrate the continued viability of this classic image processing technique. Its comparative speed and simplicity make it suitable as a prefilter to more robust techniques. We apply a large (twenty scales) multiscale Frangi filter on a subset of 201 placental images from a private database provided by the National Children's Study (NCS). We then compare several approaches of merging the multiscale result into an approximation of the PCSVN and compare them to manual tracings of the network. Finally, we develop the notion of the signed Frangi filter, upon which we describe a straightforward segmentation method called "trough-filling" which is rather suitable for this particular image domain.
|Commitee:||von Brecht, James, Ziemer, William|
|School:||California State University, Long Beach|
|Department:||Mathematics and Statistics|
|School Location:||United States -- California|
|Source:||MAI 81/1(E), Masters Abstracts International|
|Subjects:||Applied Mathematics, Medical imaging|
|Keywords:||Computer vision, Differential geometry, Image processing, Image segmentation, Placenta, Ridge detection|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be