正则化有序子集最大期望值重建算法对^(18)F-脱氧葡萄糖PET/CT图像质量和标准摄取值的影响

The effects of reconstruction algorithm of regularized ordered OSEM on image quality and SUV of 18F-FDG PET/CT

目的:探讨正则化有序子集最大期望值(OSEM)(正则化OSEM)迭代算法对^(18)F-氟代脱氧葡萄糖(^(18)F-FDG)正电子发射断层显像/计算机断层显像(PET/CT)图像质量和标准摄取值(SUV)的影响。方法:选取医院核医学科行^(18)F-FDG PET/CT检查的144例肿瘤复查患者的PET图像,图像重建采用OSEM迭代法、OSEM+点扩散函数(PSF)、OSEM+飞行时间(TOF)、OSEM+PSF+TOF和正则化OSEM(依次为M1、M2、M3、M4和M5)。测量病灶最大SUV(SUV_(max))、病灶平均SUV(SUV_(mean))、肝脏平均SUV及肝脏SUV标准差,计算信噪比(SNR)、对比度(CONT)及肝脏变异系数(CV),并比较M5算法与M1~M4算法的SUV_(max)差异性(△SUV%)。行Kruskal-Wallis秩和检验分析不同重建算法间各项指标差异。结果:M1~M5算法所得病灶SUV_(max)、SUV_(mean)总体分别呈递增趋势,M1~M5算法所得SUV_(max)、SUV_(mean)的差异均有统计学意义(H=168.0,H=152.9;P<0.05);将M5算法与M1~M4算法所得SUV_(max)、SUV_(mean)分别两两进行比较,差异具均有统计学意义(H_(SUVmax)=357.3,H=345.2,H=274.4,H=210.9;H_(SUVmean)=340.7,H=332.6,H=254.2,H=205.1;P<0.05),所有病灶△SUV%均>0(范围为2.1%~261.1%)。对于小病灶图像,M5算法较M1~M4算法所得SUV_(max)、SUV_(mean)均有明显提升,差异有统计学意义(H=37.5,H=36.7;P<0.05);对于肥胖患者的图像,M1~M5算法的SUV_(max)差异无统计学差异,SUV_(mean)差异具有统计学意义(H=10.4,P<0.05)。M5算法所得SNR和CONT较M1~M4算法差异均有统计学意义(H_(SNR1)=391.8,H=304.0,H=376.8,H=242.0;H_(CONT)=335.8,H=321.7,H=260.3,H=198.2;P<0.05)。M5算法所得CV与M1算法、M3算法和M4算法两两比较差异均有统计学意义(H=-289.1,H=-466.2,H=-218.5;P<0.05)。结论:正则化OSEM重建算法可改善PET图像质量,对病灶SUV有显著增益作用,对小病灶及肥胖患者同样适用。

Objective:To investigate the effects of iterative algorithm of regularized ordered subset expectation maximization(regularized OSEM)on the image quality and standard uptake value(SUV)of^(18)F-deoxyglucose positron emission tomography/computed tomography(^(18)F-FDG)PET/CT.Methods:The PET images of 144 patients who underwent reexamination for tumor by using^(18)F-FDG PET/CT in the nuclear department of hospital were selected.And OSEM iterative algorithm,OSEM+point spread function(PSF),OSEM+time of flight(TOF),OSEM+PSF+TOF,and regularized OSEM(successive sequence from M1 to M5)were adopted to reconstruct image.The maximum SUV(SUV_(max))and mean SUV(SUV_(mean))of lesion,mean SUV of liver and standard deviation of liver SUV were measured.And signal-to-noise ratio(SNR),contrast(CONT)and coefficient of variation(CV)of liver were calculated.And the otherness of SUV_(max)(?SUV%)between M5 and M1-M4 was compared.Kruskal-Wallis rank sum test was conducted to analyze the differences of various indicators among different reconstruction algorithms.Results:The SUV_(max)and SUV_(mean)of lesions obtained from M1 to M5 both showed general increasing trend,respectively,and the differences of SUV_(max)and SUV_(mean)among M1-M5 were statistical significant(H=168.0,H=152.9,P<0.05).The SUV_(max)and SUV_(mean)obtained from M5 and M1-M4 were respectively compared in pairs,the differences of SUV_(max)and SUV_(mean)between any two algorithms were significant(H_(SUVmax)=357.3,H=345.2,H=274.4,H=210.9,H_(SUVmean)=340.7,H=332.6,H=254.2,H=205.1,P<0.05),and?SUV%of all lesions were larger than 0(range was from 2.1%to 261.1%).For the images of small lesions,the SUV_(max)and SUV_(mean)of M5 algorithm were significantly higher than those of M1-M4 algorithm,(H=37.5,H=36.7,P<0.05).For the images of obese patients,there was no statistical difference in SUV_(max)among M1-M5 algorithm,but there was significant difference in SUV_(mean)among these algorithm(H=10.4,P<0.05).The differences of SNR and CONT between M5 algorithm and M1-M4 algorithm were significant(H_(SNR1)=391.8,H=304.0,H=376.8,H=242.0,H_(CONT)=335.8,H=321.7,H=260.3,H=198.2,P<0.05).The differences of CV among M5,M1,M3 and M4 algorithms in pairs were significant(H=-289.1,H=-466.2,H=-218.5,P<0.05).Conclusion:The regularized OSEM reconstruction algorithm can improve the PET images quality and has significant potentiation on the SUV of lesions,and it is also applicable to small lesions and obese patients.