Segmentierung und Analyse anatomischer Strukturen mit wissensbasierten Modellen

Segmentation and analysis of anatomical structures with knowledge based models. The rapid development of medical imaging modalities has lead to an increasing amount of image data, to be evaluated by the physicians. Thus it is desirable, especially in three dimensional imaging, to support the diagnostic process with fully automatic object detection and analysis. Segmenting arbitrary anatomical structures is difficult due to the sheer size of the datasets and the complexity and variability of the shapes of interest. A wide variety of deformable models has been developed, to face the versatile tasks. Neverthelessit lacks of general approaches. In reality a vast number of solutions concentrates on particular problems. Within the scope of this thesis a new knowledge based object model has been developed. Major advantage of the model is the incorporation of knowledge about image structures as well as shape variation into an overall object model. The approach is able to model two dimensional shapes as well as three dimensional surfaces. In this connection a new solution of the three dimensional correspondence problem for surfaces of arbitrary topology is presented. The relevant information is semi automatically extracted from a set of training datasets with an enhanced active contour model. Additionally, significant features on the objects surface can be defined with high accuracy. The model is fully automatic adjusted to the relevant anatomical structures using scale space methods combined with statistical optimization schemes like simulated annealing and genetic algorithms. A comprehensive validation of the model on real world and artificial data sets was also performed. Lateral radiographies of the cervical- and lumbar spine as well as three dimensional CT-datasets of the spleen and the left ventricle of the heart were used. Accurate, repeatable and quantitative geometric measures of diagnostic relevance were efficiently extracted in the subsequent analysis of the resulting shapes. Finally as an objective criterion, the segmentation quality of the model was evaluated on synthetic test images. The knowledge based object model was successfully applied on different segmentation and analysis tasks. Concerning the overall detection accuracy in the different applications the presented approach was able to overcome the limitations of traditional image processing techniques. Altogether a new deformable model is available, which provides a reliable segmentation and analysis on a wide variety of different anatomical structures in medical imaging.

Grantor dc:publisher

Publikationsserver der RWTH Aachen University

Year dc:date

2004

Author dc:creator

• Kohnen, Michael

Contributors dc:contributor

• Oberschelp, Walter

• info:eu-repo/classification/ddc/610
• Medizin
• Segmentierung
• Analyse
• Milz
• Herz
• Wirbelsäule
• Segmentation
• Active Shape Model
• Snake
• Analysis
• Spine
• Spleen
• Heart

Statement dc:rights

• info:eu-repo/semantics/openAccess

Language dc:language

ger

Identifier

https://publications.rwth-aachen.de/search?p=id:%22RWTH-CONV-123629%22

OAI identifier oai:identifier

oai:publications.rwth-aachen.de:62031