兆伏级放疗射野图像与模拟图像自动配准算法研究

目的 利用计算机图像处理与互信息比较技术,探索一种精确快速的兆伏(MV)级放疗射野图像与模拟图像自动配准算法和摆位误差分析算法,为开发自动图像引导放疗软件提供基础.方法 采用放疗患者MV级射野图像验证片,以改进滤波算法去噪、以基于偏导数阈值的灰度变换增强图像后,进行突出骨性结构同时抑制软组织和空腔的图像预处理.采用结合小波多分辨率分析的粒子群和鲍威尔混合改进算法进行互信息的参数优化和变换,与计划设计的数字重建片或X线模拟机定位片配准.以仿真人模体模拟摆位误差方法对配准算法进行验证和评估.结果 改进的图像预处理算法能很好满足MV级射野图像的骨性结构增强要求.建立的互信息配准方法兼顾了配准的精度和速度,对头颈部射野图像的自动配准过程过程平均耗时31.4 s.20例仿真体模摆位的射野图像配准验证结果显示自动配准的水平、垂直和旋转误差相对于手工配准分别降低了62.74%、67.32%和66.61%.结论 建立了基于MV级放疗射野图像的自动精确配准和误差分析算法,其配准精度和速度可以满足临床需要.

Study on registration algorithm for portal images and simulation images in megavolt radiotherapy

Objective To explore a fast and precise registration algorithm for megavolt (MV) portal images(PIs) used for radiotherapy positioning verification, and find auto analysis method of set-up error using the computed image processing and mutual information comparison technology, which provide a basis for the development of automatic image guidance software. Methods MV PIs of patients undergoing radiotherapy were tested, pre-processed with noise reduction technique based on improved filtering algorithm and contrasted by gray-scale transforming using partial derivative threshold. The bone structures were then highlighted but soft tissues and the cavities were restrained simultaneously. Improved particle swarm optimization and powell hybrid algorithm were used to optimize and transform the mutual information based on wavelet multiresolution analysis when registering the Pls with digital reconstructed radiographs (DRRs) of treatment planning or X-ray simulation-film images(SIs). Application of the designed registration algorithm was verified and evaluated through simulated set-up shifts of head and neck phantom. Results The improved noise reduction algorithm satisfactorily met the requirements for contrast of bony structures in the MV PIs. The established mutual information registration method well behaved in both accuracy and speed of registration calculation. The processing of automatic registration took only 31.4 seconds averagely for the PIs and X-ray Sis of head-neck phantom. Mean errors of automatic registration of PIs and X-ray Sis in horizontal, vertical and rotational reduced by 62. 74% ,67. 32% and 66. 61% respectively compared with manual registration in the testing of 20-cases head and neck phantom. Conclusions A precise image registration algorithm and set-up error analysis method based on MV portal images is established, and it can meet the clinical application in registration accuracy and speed.