Monomodale und multimodale Registrierung von autoradiographischen und histologischen Bilddaten

Image registration is a fundamental task in medical imaging to align two or more images. The images might be acquired at various times, by different sensors or with distinct perspectives. Medical image registration has a wide range of potential applications such as combining information from multiple imaging modalities, comparing images or reconstructing a three-dimensional volume from a series of two-dimensional images. Image registration of a series of dual-isotopic and histologic two-dimensional image data displaying slices of the brain improves tumour diagnostics and lead to an advanced understanding of the brain functionality on a molecular level. Registration is the iterative search for a transformation (rigid and non-rigid), mapping points from one image to corresponding points or structures in the second image. The aim of the registration process is the alignment of the objects contained in the images. An intrinsic registration is based on the image intensities and therefore optimises a measure of similarity calculated on the image pixels. This paper deals with the implementation and analysis of a configurable software aligning series of dual-isotopic and histologic two-dimensional images. The gradient descent and the Downhill Simplex optimisation scheme are used in combination with different measures of similarity and interpolation methods to align both monomodal and multimodal data. Available measures of similarity are the mean squares measure, normalized correlation, mutual information implemented by Mattes and normalized mutual information. The software provides the interpolation methods nearest neighbour and bspline interpolation with varying order. The conclusion shows that the implemented registration software can be executed with all measures of similarity except for normalized mutual information combined with the gradient descent and with all measures of similarity combined with the Downhill Simplex optimisation scheme. Both the monomodal and the multimodal registration of the available data show good results for all applicable combinations