Elastic registration of MRI scans using fast DCT

Analysis and comparison of functional SPECT brain scans from different subjects require a common space of study. For this purpose, an elastic matching between a MRI of reference and the MRI of the patient can be performed. A variant of a voxel-based method initially created in SPM software was implemented. A deformation map is defined as a linear combination of basis functions of the discrete cosine transform. The coefficients of this combination are the parameters of the transformation. They are optimized to minimize a cost function which represents the distance between the reference and the deformed volume. A multi-parameters and multi-resolutions scheme, combined with a conjugate gradient method, completes the optimization. A fast discrete cosine transform is used for efficient computations of the cost function and its gradient. This algorithm has been tested with 5 MRI volumes of 1283 voxels, elastically matched on a sixth MRI volume and the results show improvement upon simple nine parameters registration