Nonrigid Medical Image Registration using Adaptive Gradient Optimizer

Medical image registration has a significant role in several applications. It sequential processes, including transformation, similarity metric calculation, diffusion regularization, and optimization of the transformation parameters (i.e., rotation, translation, shear). The process for determining optimal set vectors is considered main stage affecting performance process. Hence, medical can be deliberated as an problem computing geometric transformations to realize maximum between moving fixed one. In this work, mono-modal nonrigid using B-spline designed alignment Computed Tomography (CT) images thorax Adaptive Gradient algorithm (AdaGrad) optimizer. addition, comparative study with other first order optimizers, such Stochastic Descent (SGD), Moment Estimation (Adam) (AdaMaX), AdamP, RangerQH were conducted. Also, comparison limited memory Broyden-Fletcher-Goldfarb-Shannon (LBFGS) second optimizer was also carried out. results showed superiority AdaGrad by 56.99% 48.37% improvement compared target error (TRE) SGD, LBFGS optimizer, respectively.