My New Metric
Post date: Jul 07, 2009 2:52:23 AM
I have implemented the first iteration of my novel metric that i originally discussed in my proposal. This iteration simply combines the best gradient based method that i have been using (Gradient Correlation) and the best statistical based information thoeretic metric (Mutual Information). Here are the individual results from a sensitivty analysis of GC and MI:
These plot were generated by moving the CT volume through a range of each of the 6 pose parameters, one at a time (X translation, Y translation, Z, translation, X rotation, Y rotation, Z rotation). The true optimum should be at 0 for all parameters based on the gold-standard data. I chose the GC metric from among the other gradient based methods because it had the largest capture region, was fairly smooth, and had a deep minimum at the true solution.
Here are the results from a initial guess about 4mm and 6 deg away from the true pose: I am pretty happy with these results so far.