A Variational Approach for the Segmentation of the Left Ventricle in MR Cardiac Images

Nikos Paragios
Siemens Corporate Research
Imaging and Vizualization Department

This talk proposes a new front propagation method to segment
MR cardiac images. This framework is based on the Geodesic
Active Region Model, refers to a coupled propagation of two curves (inner and outer cardiac contours) and integrates
boundary and region-based segmentation modules. The boundary
information is introduced to the objective function using the gradient vector flow framework while the region information using continuous probability density functions. The defined objective function is minimized using a gradient descent method
and the obtained motion equations are implemented using a level set approach. A recently introduced numerical approximation scheme with fast convergence rate and stable behavior is used to implement the level set motion equations. Finally, according to the application the propagations of the level set contours are
coupled using their relative distances. Encouraging xperimental results are provided using real data.


Back to Imaging in Medicine and Neurosciences