Title: Fast and accurate automatic localization of anatomical landmarks on medical images
Other Titles: Fast and Accurate Automatic Localization of Anatomical Landmarks on Medical Images
Language: English
Authors: Trapp, Martin 
Qualification level: Diploma
Advisor: Sablatnig, Robert  
Issue Date: 2013
Number of Pages: 95
Qualification level: Diploma
Abstract: 
Methods for medical image acquisition have rapidly evolved and the amount of digital images acquired in the daily image production of hospitals has exponentially increased in the last 30 years. Therefore, methods for the efficient automatic localization of anatomical landmarks on medical images need to be elaborated. Recent publications addressing this problem use regression models. Despite convenient results the main shortcomings of most of these models are superfluous time and memory consuming computations. Inspired by the memory efficient random regression fern model, the aim of this thesis is to develop a novel regression model that allows to obtain accurate results with memory efficient computations. The accuracy of this new approach is evaluated using K-fold cross validation on CT head scans, MRI T1 weighted head scans and CT whole body scans. The contribution of this thesis to the improvement of methods for the efficient automatic localization of anatomical landmarks on medical images is three-fold: (1) Two novel feature descriptors tailored to medical images are designed. One of the introduced image features (cuboidalBRIEF) outperforms all other tested feature descriptors. (2) A robust boosted regression model inspired random regression ferns is developed. The model stands out through its significantly higher accuracy as well as time and memory efficient computations. (3) A generalized multi-phase landmark location system allowing is presented. While the second phase results turn out to be less accurate than anticipated the first phase results of the system are highly satisfying.
URI: https://resolver.obvsg.at/urn:nbn:at:at-ubtuw:1-62039
http://hdl.handle.net/20.500.12708/2545
Library ID: AC11320828
Organisation: E183 - Institut für Rechnergestützte Automation 
Publication Type: Thesis
Hochschulschrift
Appears in Collections:Thesis

Files in this item:


Page view(s)

12
checked on Jul 30, 2021

Download(s)

57
checked on Jul 30, 2021

Google ScholarTM

Check


Items in reposiTUm are protected by copyright, with all rights reserved, unless otherwise indicated.