非刚性医学图像的博弈配准方法

Nonrigid medical image registration approach based on game theory

提出一种通过像素间的非合作博弈实现非刚性医学图像配准的方法。首先用离散位移场表示形变,将配准问题归结为马尔可夫随机场中的最大后验概率推理问题,并转化为吉布斯随机场的能量极小问题。然后将配准视为像素间的博弈,通过寻找纳什均衡点获得能量极小点,完成图像配准。博弈过程中,各个像素都选择使自身收益最大化的位移向量。该方法能配准任意形变,需要设定的参数少,易于实现。实验结果表明,博弈配准方法的配准能力强,精度高,对图像分辨率的变化具有鲁棒性。

A nonrigid medical image registration approach based on noncooperative game among pixels is proposed.First,deformations are represented as a discrete displacement field,thus registration problems can be considered as maximum a posteriori inference problems in Markov random field and optimized using pairwise Gibbs energy minimization technique.Then,registration process is modeled as a game among pixels,and Gibbs energy function is optimized by finding a Nash equilibrium of the game.During the game,each pixel selects a displacement to maximize its payoff.The game-theoretic approach can register any deformation with high accuracy and a few parameters.Experiment results show that the proposed approach is robust to resolution variations.