Optimization of contour detection in cardiac MRI using Active Appearance Models: definition of the training set
van der Geest R, Reiber J, Lelieveldt B, Matzken S, Angeli e
Active appearance models (AAMs) have been successfully used for image segmentation in cardiac MR (1). But its availability in clinical routine remains still low. This work aims at defining the optimal training set used in the AAM to bring it closer to the clinical routine: The minimal number of images to be included and the characteristics of the images were our focuses. We concluded that a minimal number of 150 images (mixture of 80% normal and 20% pathological) should define the training set.