الفهرس 
Matching-based registration
Methodology of the Proposed MedicalOnly 14 pages are availabe for public view
 |  | from | 106 |  |  |
| | | from | 106 | | |
Abstract
Medical image registration (MIR) process has a significant role in different medical applications. It is used for aligning two images by determining the optimal transformation parameters of the control points of the transformed image, which connect the moving image (MI) and the fixed image (FI). The general framework of the different MIR methods includes transformation, the similarity factor, and the optimization technique. Typically, image registration can be considered an optimization-based process that uses an objective function to measure the similarity between the FI and MI to identify the optimal transformation parameters. The registered images include more information than the moving image, which aids the clinicians in identifying and finding disorders in the organs. In this thesis, two models are proposed to provide an efficient MIR system based on selecting the proposer transformation and optimization methods. Model 1 is based on non-parametric transformation. On the other hand, Model 2 is based on the B-spline transformation, and these two models are based on the AdaGrad optimizer by using the Mean Square Error (MSE) or Normalization Correlation Coefficient (NCC) as an objective function to obtain the optimal transformation parameters. Moreover, comparative studies were carried out using different first order optimizers, such as Stochastic Gradient Descent (SGD), Adaptive Moment Estimation algorithm (AdaMaX), AdamP, RangerQH, and the limited memory Broyden-Fletcher-
iv.