图像重建算法对兆伏级成像系统图像质量及自动配准精度的影响研究

Study on the influence of reconstruction algorithms of image on the image quality and precision of automatic registration of imaging system with megavolt grade

目的:对比研究TOMO Radixact螺旋断层放射治疗系统不同迭代重建(IR)算法对兆伏级CT(MVCT)成像系统图像质量的改善及对自动配准精度的影响,探索适合临床应用的重建算法。方法:使用MVCT成像系统分别扫描Tomo-Phantom HE型模体和Catphan604型模体,分析MVCT图像重建算法中的标准(STD)算法、迭代重建通用(IR-G)算法、迭代重建软组织(IR-ST)算法3种重建算法生成的3组图像,分别计算3组图像的噪声指数(NI)、图像均匀性指数(UI)、调制传递函数(MTF)、低对比度可见度(LCV)和对比度噪声比(CNR)。利用Lucy模体进行自动配准算法准确性测试,通过对左右方向(x轴)、头脚方向(y轴)、垂直方向(z轴)以及y轴的自由旋转(Roll)4个维度的配准数据进行其对自动配准精度的影响验证分析。结果:IR-G和IR-ST重建算法的NI分别为39.58±0.10和14.62±0.26,优于STD算法的39.58±0.10,UI分别为19.87±0.83和15.84±2.51,优于STD算法的24.51±1.81,LCV分别为2.50±0.03和1.74±0.11,优于STD算法的3.67±0.04,且均有明显提升,但MTF50%的高对比度分辨率分别为0.23和0.21lp/mm,低于STD的0.32lp/mm,整体图像质量IR重建算法优于STD重建算法。自动配准算法准确性测试中,y轴方向上IR-G和IR-ST算法的配准精度分别为(0.360±0.142)mm和(0.245±0.050)mm,均高于STD算法的(0.145±0.136)mm,差异有统计学意义(Z=6.0、15.0,P<0.05),其余方向的配准精度差异均无统计学意义(P>0.05)。结论:IR算法在噪声、均匀性及低对比度分辨率上均有优势,但高对比度分辨率有所降低,高对比度分辨率的降低并未导致自动配准精度降低。

Objective:To compare and study the improvement of different iterative reconstruction(IR)algorithms of the tomotherapy(TOMO)Radixact system on the image quality of megavoltage computed tomography(MVCT)imaging system,and the influence of that on the precision of automatic registration,and to explore the reconstruction algorithm that is suitable for clinical application.Methods:Using the MVCT imaging system to respectively scan the Tomo-Phantom HE phantom and the Catphan 604 phantom,and to analyze three groups of images were generated by three kinds of reconstruction algorithms,which included the Standard(STD)algorithm,IR General(IR-G)algorithm and IR Soft Tissue(IR-ST)algorithm,in MVCT image.The noise index(NI),uniformity index(UI)of image,modulation transfer function(MTF),low contrast visibility(LCV)index and contrast-to-noise ratio(CNR)of three groups of images were calculated respectively.The Lucy phantom was used to test the accuracy of automatic registration algorithm.The registration data of four dimensions,included left-right(X-axis),head-foot(Y-axis),vertical(Z-axis)and free rotation(Roll)around the Y-axis,were used to conduct verification analysis for the influences of them on the precision of automatic registration.Results:The NI values of IR-G and IR-ST reconstruction algorithms were respectively 39.58±0.10 and 14.62±0.26,which were better than 39.58±0.10 of STD algorithm,and the UI values of them were respectively 19.87±0.83 and 15.84±2.51,which were better than 24.51±1.81 of STD algorithm,and LCV values of them were respectively 2.50±0.03 and 1.74±0.11,which were better than 3.67±0.04 of STD algorithm.All of them appeared significant increase,but the resolution with high contrast of MTF were respectively 0.23 and 0.21,which were lower than 0.32 of STD.The overall image quality of the IR algorithms was superior to that of the STD algorithm.In the accuracy test of the automatic registration algorithm,the registration precisions of IR-G and IR-ST algorithms on Y-axis were respectively(0.360±0.142)mm and(0.245±0.050)mm,which were significantly higher than 0.145±0.136 of STD algorithm,and the differences of them were significant(Z=6.0,15.0,P<0.05).The differences of registration precisions of other directions were not significant(P>0.05).Conclusion:The IR algorithm has advantages in terms of noise,uniformity and resolution with low-contrast,however,it shows reduction on resolution with high-contrast.This reduction of resolution with high-contrast do not lead to the decrease of the precision of automatic registration.