Registration Paramter Evaluation

A method for evaluating and tuning the parameters of image registration algorithms has been developed using Gabor wavelets. The filter has been adapted as a 3D local anatomical structure descriptor, named the Maximum Responded Gabor Wavelet (MRGW), and measures registration quality based on anatomical variability of registrered images. Through testing on a dataset of T1 brain volumes, the entropy of MRGW was successfully applied for the fine-tuning of parameters of a nonlinear spatial normalization algorithm. Using the parameters selected through this framework, the normalization algorithm was shown to achieve less anatomical variability among data. This is clearly indicative of the generalization properties of the parameter tuning method. Additional benefits of this algorithm have been found when compared with similar parameter tuning approaches. With regards to computational complexity, the runtime scales linearly with number of image volumes. The segmentation process found in prior methods is not required in the MRGW approach.

Citations: [·]
Last updated 2014 | School of Computer Science