MEDx Frequently Asked Questions - Registration

Q: What is the Initial Sampling parameter in AIR?

The "Initial Sampling" parameter decides as to how densely you sample your data. The larger the number the fewer the pixels that are sampled. In memory, the image data is stored in a 1D pointer, so the entire 3D volume is kept in memory as a 1D array. Every 81st (or whatever number you use; 81 is used because it's an even power of 3) number in this 1D array is then used in the calculations. In the next pass, every 81/3=27th number is used; then, every 27/3=9th, then every 9/3=3rd, then finally every voxel is used.

This is the reason why Roger Woods recommends that the sampling not have factors in common with the image dimensions. Here's an excerpt from the official web page,(http://bishopw.loni.ucla.edu/AIR3/alignlinear.html#initial_sampling):

initial-sampling -- controls how densely data is sampled during the first iterative cycle of the algorithm. Large values generally speed up the registration process because gross misregistration can be detected with fairly superficial sampling of the data. However, choosing an excessively large value can be counterproductive if the algorithm falls into an infinite loop or is led far from the true value by nonrepresentative sampling. Avoid multiples of two when choosing sampling parameters. If any of your matrix dimensions are divisible by two, the sampling will become spatially biased until the sampling density reaches one, at which point the algorithm will have to iteratively overcome the earlier bias at the maximal sampling density. If your matrix dimensions are divisible by three, you will have a similar problem with sampling densities that are multiples of three.

Other FAQ Topics

If you have a question that you would like to see addressed in our list of Frequently Asked Questions, please contact customer support.

Back to FAQ Home