An Efficient Medical Image Registration Algorithm Based on Gradient Descent

Image registration is a key technique in medical image analysis. Lucas-Kanade algorithm is a high precise optimization algorithm in correlation methods. However, huge computational cost prevents it from application. Then an efficient medical image registration algorithm based on gradient descent is introduced in this paper. Firstly, the algorithm redefines the objective function by switching the role of the image and the template. Then, the Gauss-Newton gradient descent algorithm is used to get the increments of the parameter. This ensures the Hessian to be constant in every step of iteration and can be pre-computed. At last, the parameter is iteratively solved until it satisfied the condition of convergence. Experimental results with several standard medical sequences, which are using the set of affine warp, show the improved algorithm ensures the precision and improve the computational efficiency.