Shape-Based Analysis of Cardiac Images: 3-D Active Appearance 
Models for MR, CT, Echo, and Angiography Segmentation


Milan Sonka
Professor of Electrical and Computer Engineering
The University of Iowa, Iowa City IA 52242
e-mail: milan-sonka@uiowa.edu
http://www.icaen.uiowa.edu/~sonka


Abstract:
Traditional computerized approaches to medical image segmentation 
and analysis are either border based or region based. Border 
based segmentation methods only rarely consider object shapes 
and do not take region-based properties into account . Region 
based segmentations assume some kind of homogeneity of the object 
and rarely utilize information about object shape. The assumptions 
of the traditional methods are frequently violated in medical 
image segmentations due to natural anatomical variability. Point 
distribution models (PDMs) and even to a larger extent the 
recently  introduced active appearance models (AAMs) allow to 
incorporate  shape and appearance information simultaneously 
and are thus well suited for medical image analysis.

Several medical image analysis applications will be described in 
which PDMs and AAMs helped to achieve breakthrough performance. 
Improvements to the PDM and AAM fitting processes that further 
increase the applicability of active model approaches will be 
discussed. Applications include 2D, 2D + time, and 3D cardiac 
MR, CT, angiography, and cardiac echo  image segmentation. 
Assessment of heart function is the primary goal in the cardiac 
applications. Comparison of individual approaches in clinical 
applications, reproducibility and accuracy of individual methods 
will be given and demonstrated on clinical examples.