A cluster-assisted global optimization method for high resolution medical image registration
Rongkai Zhao, Geneva G. Belford, Mohamed Taha Gabriel · 2004
Optimization is a key component of image registration. Due to the non-convexity and high computation cost of the objective function, a common tactic is to set an initial guess and then use multi-resolution or local optimization methods to find a local optimum of the objective function. For almost all local optimization methods, the initial location in the search space plays a critical role in the accuracy of the registration. Initial guesses are often obtained through data-specific methods. The paper offers a new hybrid optimization method assisted by a density-based clustering algorithm. The new method is less data-specific and more suitable for semi-automatic or automatic image registration. Global optimization does not guarantee timely convergence. A genetic algorithm is a component of our hybrid method; however, our method usually converges within a reasonable time. This new method has been applied to registering high resolution brain images.