Optimized monomodal image registration using cuckoo search algorithm

• Main Author: Md. Roslan, Muhammad Syafiq
• Format: Thesis
• Language: English
• Published: 2022
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/101592/1/MuhammadSyafiqMSC2021.pdf; http://eprints.utm.my/id/eprint/101592/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:150714

Medical image registration, which is employed in analyzing the similarity merits in helping the diagnosis is an important part of the medical image analysis. The process involves combining two or more images in order to provide more information. Therefore, there is a need for a method that can produce an image as a registration result that can produce more information without any loss of the input information and without any redundancy. The accuracy and computation time of the existing picture registration approach are now in question, although they could be improved if an optimization methodology is applied. Hence, this research proposed an enhancement of the image registration process focusing on monomodal registration by incorporating an optimization method called Cuckoo Search (CS) algorithm with Levy flight generation. This method was used to find the optimum parameter value (Gradient Magnitude Tolerance, Minimum Step Length, Maximum Step Length) and it was tested to brain, breast and kidney cancer that are captured on Magnetic Resonance Imaging (MRI) image. The performance of the proposed method was then compared with standard monomodal registration. For all the cases investigated, the experimental results were validated by measuring the following: Mutual Information (MI), Normalized Mutual Information (NMI), Mean Square Error (MSE), Coefficient Correlation (CC) and Central Processing Unit run-time. The results of the study illustrated that the proposed method achieved the best 2% improvement in MI, NMI, MSE, CC results. In addition, the proposed method reduced about 40% in Central Processing Unit run-time as compared to the benchmarks methods. This indicates that the proposed method has the potential to provide faster and better medical image registration results.