Multi-objective Evolutionary Registration of Medical Images with the Direct Piece-Wise Transformation in Image Response Space

The paper presents a method of multi-objective piece-wise evolutionary registration of medical images. The evolutionary search is transferred from the actual image into response space, which accounts for the reliable extraction of the main shape features while avoiding the direct comparison of the pixel values. A composite two-layer image model is constructed whereby the first layer outlines the principal shape of the image while the second layer refines the model with its piece-wise representation. The evolutionary search for the optimal image mapping works in two consecutive phases. The global optimization phase seeks for the global components of the principal shape transformation which later serves as the initial approximation in the search for the final optimal piece-wise mapping. The algorithm works with multiple populations and uses several fitness functions as multiple optimization objectives. Computational experiments show that the proposed method is capable of finding the global components of the template transformations as well as their local deformation.