Multi-modal image registration - overview
Given are two three dimensional images R and T. One is a "Magnetic Resonance Image" (MR) of an human brain and the other one is the "Atlas of the Human brain". This atlas contains drawings of a coronal sliced brain of a different human brain.
Our task is to find a transformation on the template T to match it on the reference image R. In our case we choose the atlas for T and the MR for R. Since colors of both images are completly different, the algorithm cannot match them concerning the colors itself. Both images contain the same structures, hence one can match on morphologic information, e.g. contour lines. T will be elastically deformed to match contour lines of R. As soon as the transformation is calculated out, it can be applied on each of the three color channels of the colored Atlas, which results in a colored atlas matching the MR brain. Because Brain structures are marked with specified colors, the matched atlas offers detailed information on the MR areas to medical scientists.
Remark: Due to the fact that the cerebellum is not drawn in the atlas yet, it is cut off in the MR as well.
Remark: With examples starting by "v146" coronal slices of the atlas are adjusted to each other by e144_02 and since "v149" template is adjusted by e149_02. The latter is done with a shape preserving bicubic algorithm. Tricubic interpolation is done analog. A comparision of original and adjusted atlas slices is available, as well.
- v146_03: r=[4, 3.5, 3], β=2, hist=32, interp=lin (adequate rigidity)
- v146_04: r=[4, 3, 2.5], β=2, hist=32, interp=lin (low rigidity)
- v146_05: r=[4, 3.5, 3], β=1, hist=32, interp=lin (low volume conservation)
- v154_01: r=[2.5, 2.5, 2.5, 2.5], β=0.5, hist=16, interp=tricub, measure=1.2.1
- v154_02: r=[2.5, 2.5, 2.5, 2.5], β=0.5, hist=16, interp=tricub, measure=1.3.1 (second-best)
- v156_04: r=[3,3,3,3], β=0.5, hist=16, interp=tricub, measure=1.3.1, reg_alpha=10-3, Ustart indirect 1/2. (like v156_05, but here gray=(R+G+B)/3)
- v156_05: r=[3,3,3,3], β=0.5, hist=16, interp=tricub, measure=1.3.1, reg_alpha=10-3, Ustart indirect 1/2. (smooth displacement and good fit: best by now)
- v156_06: r=[2.5,2.5,2.5,2.5], β=0.5, hist=16, interp=tricub, measure=1.3.1, reg_alpha=10-3, Ustart indirect 1/2.
- v156_07: r=[2.5,2.5,2.5,2.5], β=0.5, hist=16, interp=tricub, measure=1.3.1, reg_alpha=10-3, Ustart indirect 1/2. (like v156_06, but here gray=(R+G+B)/3)