【摘要】 目的对多模态非刚性变换序列图像进行配准。方法将一种新的信号处理的概念引入配准过程,以两组具有时延特性的随机信号分别描述待配准的两幅医学图像的边缘特性,继而提出一种以信号互相关函数为性能指标,通过利用神经网络的泛化能力对轮廓特征点样本进行训练以得到最优变换参数的头部断层扫描图像自动配准算法。结果仿真结果表明该算法配准误差可达到亚象素级以下,且比之其他基于形状信息的配准算法具有寻优参数少,配准时间短,自动化程度高的特点。最后该算法被成功地应用到了做过开颅手术病人的CT-MRI图像融合上。结论该方法为多模态医学图像配准提供了一种新的有效手段。更多还原

【Abstract】 Objective To co-register multi-modal serial images by non-rigid transforming. Method A new conception of signal process was introduced to the procedure of medical image registration. The edge of two frames of medical images as two rows of random signals that have time delay characteristics was described. With the correlation function of the signal as the measure, the transform relationship between the two images was optimized by means of an artificial neural network. This method was successfully developed for brain image co-registration. Result Computer simulations were conducted and the simulation results demonstrated that the co-registration error was smaller than one pixel. Furthermore, the present method had fewer parameters to be optimized, less time consumed and were more automatical than other co-registration methods. Finally, it was demonstrated that the present method can successfully co-register the post-operative CT images with the pre-operative MRI images in a patient’s undergoing neurosurgical operation. Conclusion This method provides a new useful tool for multi-modal medical images co-registration.更多还原