next up previous contents
Next: Evaluation Up: Deformable Templates Previous: Getting started   Contents

Practical

In this practical we shall be assessing the performance of the deformable template when presented with the task of tracking the outline of the descending aorta in an MR sequence of 20 images of the heart. As there is no ground truth data available for the dataset we shall concentrate on the reproducibility of the technique under changes in parameters and also against human operators.

First the model must be trained. You will have to hand outline some images in order to train the system. To do this select r axis from the markup menu and specify the axis grid. Subsequent points are selected according to the blue radial line. You will not need not to include again the bounding points which have already been specified from the axis. Change the File name base to a unique name and output the file. When done click Make and Output. Select the next image you wish to mark up (every other image will be sufficient). Reset the mark up by selecting r axis from the markup menu again and continue.

The next step is to build the model itself. To do this you need a file that lists all of the model files and their relative directory, this should have been automatically constructed and will have the extension .blt. For example if you had named you files aortamodel0, aortamodel1...aortamodel10 in the directory ./tmp then the file you create would look like this;

/tmp/aortamodel0
/tmp/aortamodel1
.
.
/tmp/aortamodel10

When this is done ensure that the File name base is set to the name of your list file and hit the Build and Output PCA button. You now have a PCA model of this dataset.

In order to use the model you should load the mean model .mml file and hit show. Load the PCA and hit search in order to locate the aorta outline. Hit show to redraw the result (note, in some instances the mean model is sufficiently different from the data that the system is unable to find the global minima. In such cases feel free to either move the mean model orchange some of the other model parameters.)


next up previous contents
Next: Evaluation Up: Deformable Templates Previous: Getting started   Contents
root 2017-11-20