Application of point distribution models to the automated analysis of echocardiograms

We have recently described how a Point Distribution Model (PDM) can be used to automatically detect structures in medical images. The method has two stages. Initially a statistical model of the shape variability is generated from a set of training images. Structures are located in experimental data by placing an instance of the model in the image and iteratively refining its shape and location to better fit the data. Whilst fitting to the image the model can only adopt shapes which are consistent with the training data. We call the combination of a PDM and the iterative refinement procedure an Active Shape Model (ASM). We have `trained´ models of the left ventricle and associated structures as seen in both long axis and short axis echocardiograms. We show that ASM search using these models is able to detect the structures of interest rapidly accurately and reliably