| Objective: To compared the practicability and accuracy of automatically segmented and measured metabolic tumor volume(MTV)in FDG avid liver tumors by different thresholds in different levels,in order to find out a suitable threshold algorithm.Methods: A total of 57 patients with 78 lesions who took PET/CT examinations in Nuclear Medicine Department of Eastern Hepatobiliary Surgical Hospital in Anting during from 04,13,2016 to 09,30,2016 were enrolled into this retrospective study in the inclusion criteria of finding FDG-avid liver tumors and the exclusion criteria based on pre-PET/CT therapies and PET/CT images.Different thresholds(Th)in different levels were used for auto-segmentation and measurement of MTVs,including percentage thresholds(thresholds=30%,35%,40%,45% or 50% of SUVmax-lesion),PERCIST thresholds(thresholds=3SUVmean-liver+2SD or thresholds=2×SUVmean-blood-pool)and source-background thresholds(SUVmean-liver+k×(SUVmax-lesion-SUVmean-liver),k=0.15,0.2,0.25,0.3,0.35 or 0.4).Optimal threshold(ThOP),optimal percentage and MTVVIS were determined on the standard that the iso-contours originated by manly decided thresholds in PET images and low attenuation boundaries in CT images matched well by visual.Matched t-test or Wilcoxon signed-rank test were separately used for statistical analysis between different thresholds and optimal threshold.χ2 test was used for statistical analysis of measurable rates of MTVs by different thresholds.Wilcoxon signed-rank tests were separately used for statistical analysis between MTVs by different thresholds and MTVVIS.Linear regression was further used if there was no significant difference on Wilcoxon signed-rank test for MTVs to analyses accuracy.Results: Th30,Th35,ThL,ThBP,Th0.15 and Th0.2 were less than ThOP(P<0.05),while Th45,Th50,Th0.3,Th0.35 and Th0.4 were higher than ThOP(P<0.05).Th40 and Th0.25 were the only two thresholds that had no significant difference with ThOP(P=0.1025、P=0.8591).MTV30,MTV35,MTV40,MTV45,MTV50,MTVL,MTVBP,MTV0.15,MTV0.2,MTV0.25,MTV0.3,MTV0.35 and MTV0.4 are successfully measured in 38,50,55,63,71、67,17,75,76,78,78 and 78 lesions separately.There was a significant difference among succeed-rates of MTVs segmented by percentage thresholds in different levels in total(χ2=51.350,P<0.0001),and the rates decreased as the percentage levels fell(linear-by-linear association=47.282,P<0.0001).Application of percentage thresholds in low percentage levels were limited in lesions with low source-to-background ratios(SBR).Normal liver tissues were often included in MTVL therefore it has a minimum segmentation rate.Segmentation rates of MTV40,MTVBP and MTV0.25,which were choose as representatives for percentage thresholds,PERCIST thresholds and source-background thresholds,were compared on adjusted significant level of 0.0125.MTVBP(χ2=11.180,P=0.004)and MTV0.25(χ2=21.061,P<0.0001)had higher rates than the rate of MTV40.There was yet significance between the segmentation rates of MTVBP and MTV0.25(χ2=7.300,P=0.026).MTV30,MTV35,MTVL,MTVBP,MTV0.15 and MTV0.2 were less than MTVVIS(P<0.05),and MTV45,MTV50,MTV0.3,MTV0.35 and MTV0.4 were overestimated compared to MTVVIS(P < 0.05).There was no significant difference compared MTV40 with MTVVIS(P=0.0632)and MTV0.25 with MTVVIS(P=0.1498),while MTV0.25 had better agreement with MTVVIS.Results: SUVmean-liver + 2SD was not a suitable thresholds algorithm for MTV measurement in liver,as that normal liver tissue could be included in by this method.Simple percentage thresholds lead to overestimated MTV in low SBR lesions and underestimated MTV in high SBR lesions.Low level percentage thresholds on low SBR lesions were also less practicable than thresholds of 2×SUVmean-blood-pool or source-background thresholds.Source-background threshold based on the algorithm of SUVmean-liver + 0.25 ×(SUVmax-lesion-SUVmean-liver)has an advantage over PERCIST threshold of 2 × SUVmean-blood-pool in accuracy.In conclusion,source-background threshold on SUVmean-liver + 0.25 ×(SUVmax-lesion-SUVmean-liver)outperforms percentage thresholds and PERCIST thresholds in practicability and accuracy for automatically segmented and measured MTV in FDG avid liver tumors. |
