Fusion Of Medical Images By Using Artificial Intelligence Models

Abstract

As a result of two distinct modalities employed in the image acquisition, patient posture and resolution of two images can differ. This causes misalignment of two images while fusing. Hence, image registration is a necessary pre-processing step prior to image fusion. Multimodal image registration avoids or reduces the above-mentioned disadvantages and enhances fusion quality. Image registration superimposes PET and MR images into one coordinate system so that the data from the same physical objects are combined together. Medical imaging is the method of visual representation of body parts of the human body. Medical imaging is carried out with the assistance of scanning modalities which are classified broadly as functional and anatomical modalities. The key aim of this thesis is to construct effective medical image fusion techniques to deliver valuable clinical information to support the medical professionals. Such techniques must fulfill some requirements. The first is to transfer the essential information contained in the input images into the fused image without losing any of it. Second requirement is to remove artifacts in the fused image. We suggest methods by taking these requirements into account as well as taking the other constraints within literature into account. The most important thing when devising our methods is to leverage the adaptive image processing algorithms. The adaptive idea ensures low information loss and small artifacts.