Qclear重建技术GE.pptx

Phantom and clinical evaluation of the Bayesian penalized likelihood reconstruction algorithm Q.Clear on an LYSO PET/CT system J Nucl Med.2015 September Eugene J Teoh Churchill Hospital,Oxford University Hospitals PET重建方法：重建方法：FBP法、法、MLEM（最大似然期望法）和（最大似然期望法）和OSEM法法FBP法法 优点：断层重建简单而快速。优点：断层重建简单而快速。缺点：缺点：ramp滤波器去除图像星状伪影高频噪声时，增加图滤波器去除图像星状伪影高频噪声时，增加图像的空间分辨率像的空间分辨率,但同时放大了噪声但同时放大了噪声,尤其是在低计数数据采集的尤其是在低计数数据采集的时侯。时侯。MLEM 优点：相对欠采样、低计数的核医学成像中可发挥其高分优点：相对欠采样、低计数的核医学成像中可发挥其高分辨的优势辨的优势缺点：运算量大、运算时间长。缺点：运算量大、运算时间长。OSEM 在在MLEM的基础上发展起来的的基础上发展起来的,增加有序子集。增加有序子集。优点：加快收敛速度，减少运算时间，提高图像质量。优点：加快收敛速度，减少运算时间，提高图像质量。缺点：产生的为非完全收敛图像，完全收敛需要增加迭代次数，缺点：产生的为非完全收敛图像，完全收敛需要增加迭代次数，会产生较大噪声。会产生较大噪声。Limitations of PET:spatial resolution(5mm)There is no agreed standardised uptake value(SUV)for differentiation of disease.This is in part because the image reconstruction methodologies used have an impact on the accuracy and reproducibility of SUV measurements.Introductionpenalized likelihood iterative PET reconstruction（Q.Clear）It includes point spread function(PSF)modelling and controls the noise through the use of a penalty term.Due to the penalty function,allows an effective convergence to be achieved in images,potentially providing a more accurate SUV.This penalty function acts as a noise suppression term and is controlled by a penalization factor(termed beta).Q.Clear has been shown to significantly improve signal-to-noise in clinical scans compared to OSEM,particularly in small faintly avid abnormalities.The aim of this study was to determine the optimum penalization factor(beta)for clinical use this was achieved by blinded clinical scoring a subset of beta values which were determined from phantom studies.A NEMA image quality phantom was the six smallest spheres filled with an activity concentration in a 4:1 ratio to the background activity concentration.Phantom(ToF)OSEM protocol(2 iterations(i)24 subsets(ss)6.4mm Gaussian filter),(ToF)OSEM PSF protocol(3i,24ss,2mm Gaussian filter)Q.Clear over a range of beta values(100,200,up to 1000).MATERIALS AND METHODSParameter measurement:Contrast recovery(CR)Background variability(BV)Residual lung error(LE)Fifteen 18F-FDG PET/CT scans were retrospectively selected.nodule evaluation(5 scans)staging of NSCLC(5 scans)staging of colorectal cancer with known liver metastases(5 scans).9men and 6 women Clinical Evaluation-Case SelectionDiscovery PET/CT 69090min3D mode skull base to upper thighsfast for at least 6 hours4 MBq/kgautomA with a noise index of 25pitch of 0.984,120 kVThe reconstruction algorithm of study:ToF OSEM(VPFX,GE Healthcare),used with 2i,24ss and 6.4mm filterToF OSEM PSF protocol(3i,24ss,2mm filter)Q.Clear reconstruction algorithm for penalization factors(betas):200,300,400 and 500Clinical Evaluation Reconstruction and AnalysesVisual analyses of the OSEM,OSEM PSF,and Q.Clear PET images,six reconstructions per case.Cases were reviewed sequentially,and the reconstructions were ranked(from 1 to 6)according to six image quality(IQ)parameters:Overall IQ(1 excellent,5 worst)Background liver IQ(1 excellent,5 worst)Background mediastinum IQ(1 excellent,5 worst)Background marrow IQ(1 excellent,5 worst)Lesion detectability(1 excellent,5 poor)Noise level(1 minimal,5 unacceptable)RESULTSOSEM PSF was the lowest ranked construction for all the liver metastases cases(Figure 4)and a majority of the mediastinal node cases(Scorer 1:3/5 cases,Scorer 2:5/5 cases)due to the degree of inherent noise in these respective organs.OSEM PSF ranked second highest for lesion detectability in a majority of the lung nodule cases(Scorer 1:5/5 cases,Scorer 2:4/5 cases)(Figure 5),which is explained by low noise levels observed in the lungs.DISCUSSIONThe phantom work was used to narrow the choice of beta values used in the clinical part of the investigation as well as to investigate the properties of the new Q.Clear reconstruction.The majority of Q.Clear reconstructions have a higher CR and lower BV than OSEM suggesting that Q.Clear is an improved reconstruction.Two reason:The effective convergence achieved in Q.ClearThe inclusion of PSF modeling into the Q.Clear algorithmWhen Q.Clear is compared to OSEM PSF some Q.Clear reconstructions have a higher CR and some lower,depending on the beta value.With the exception of B100,Q.Clear reconstructions have lower BV than OSEM PSF.There is no statistical significant difference(p0.05)between BV from OSEM PSF or B100.Increased CR(from OSEM PSF)OR Decreased BV(from Q.Clear)?Ideally these points on the graph would lie in the top left of each figure(Figure 1).OSEM PSF is not used due to the noise seen in the clinical images and so the B100 reconstruction can be discounted for the clinical part of the investigation due to its similarity to OSEM PSF.For the smallest sphere(10mm)there is a significant increase in CR for B100-400(p0.003),no significant difference for B500(p=0.3)and a significant decrease in CR for B600-1000(p0.05)OSEM PSF had higher BV than OSEM and this was significant(p OSEM (p0.003)but B500 for the smallest sphere The significances of these differences(between CR OSEM PSF and B200)for the six spheres from smallest to largest.CR was highest in the majority of spheres using B200,followed by OSEM PSF.(highest in the lung nodule cases)Considering the observed fall in CR beyond B500,this to some extent again corroborates the findings of the phantom study.Suggesting that B400 would be the most appropriate choice that strikes a balance between optimising CR and image noise levels The CNR for all spheres was higher for Q.Clear than OSEM,which was significant(p0.05)between the CNR from Q.Clear and OSEM PSF but 200 for the two smallest spheres.Q.Clear subjectively improves image quality and increases CR and decreases BV in phantom studies compared to standard OSEM reconstructions.A beta value of 400 appears to be the optimal value in 18F-FDG oncology PET/CT using the new Q.Clear reconstruction algorithm.CONCLUSION