基于支持向量回归的PET/CT图像衰减校正方法

摘要

目的 在电子发射及计算机断层扫描系统(positron emission computed tomography/X-ray computed tomography,PET/CT)图像衰减校正的能量转换过程中,为了改进双线性转换法用线性关系来拟合非线性关系的不足,本文以支持向量回归为基础,提出了一种新的能量转换法即支持向量回归的PET/CT图像衰减校正方法来进行衰减校正,以寻找CT值和511 keV能量下线性衰减系数值之间的最佳转换关系.方法 使用仿真软件GATE(Geant4 Application Tomographic Emission)模拟了11组不同材质的圆柱体体模.然后根据GATE仿真的不同材质圆柱体体模,求出其CT值和511 keV能量下线性衰减系数值并代入SVR模型中进行训练,建立CT值和511 keV能量下线性衰减系数值之间的SVR模型.最后与目前PET/CT衰减校正能量转换中常用的双线性能量转换法进行对比分析,并分别应用于GATE仿真的NCAT(NURBs Cardiac Torso)像素体模图像中,评估两种方法准确性的差异.结果 支持向量回归的PET/CT图像衰减校正方法得到的511 keV能量下对应物质的线性衰减系数值的相对百分误差值较小(肺的相对百分误差值3.1%和肝脏的相对百分误差值1.08%),且经过支持向量回归法衰减校正的PET图像,其MSE评价值都是最小的(176.9230),其PSNR和AG的评价值都是最大的(31.8621和7.9083).这说明经过支持向量回归法衰减校正的PET图像相比于双线性转换法衰减校正的PET图像,更接近于静态的图像.结论 支持向量回归的PET/CT图像衰减校正方法在PET/CT图像的衰减校正应用中有更好的表现,可以更好地吻合CT值与511 keV能量下线性衰减系数之间的转换关系,从而提高了PET/CT图像的衰减校正效果,改善了PET/CT图像定量的准确性,便于医生做出更精确的临床诊断.%Objective A new energy conversion method based on support vector regression has been put forward to enhance the function of former attenuation correction process.Support vector regression based PET/CT image attenuation correction method is used to make up the shortcomings in terms of adopting linear relation to fit nonlinear relation when using bilinear transformation method in the energy conversion process of electron emission or PET/CT image attenuation correction and find the best conversion relationship between CT value and linear attenuation coefficient that under the energy condition of 511 keV.Methods The simulation program called GATE (Geant4 Application Tomographic Emission) was adopted in this work to simulate 11 groups of cylinder phantoms in different kinds of materials.According to the data of the simulated cylinder phantoms,the CT values and the linear attenuation coefficients of them could be calculated and then get trained through SVR model to set up new models.Finally,the new method and the former one were both applied to the GATE stimulated NCAT (NURBs Cardiac-Torso) pixel phantom image analysis process to evaluate the accuracy of them separately.Results The results showed that the calculated relative error percentages of the new method were smaller under the condition of 511 keV using PET/CT image attenuation correction method (i.e.the relative error percentages for the data of the lung and liver were 3.1% and 1.08%).Additionally,the corrected PET images could have the minimum MSE value (176.9230) ,the maximum PSNG and AG value (31.8621 and 7.9083).The PET images corrected by the new method were more similar to the static images in comparison with those using former process.Conclusions The performance of the support vector regression based PET/CT image attenuation correction method is better than the previous one.The new method is more consistent with the transformational relations between CT value and the linear attenuation coefficient under the energy condition of 511keV,hence it could have better accuracy in analyzing PET/CT images quantitatively therefore enhancing the precision in clinical diagnosis as well.