¹⁸F-FDG PET/CT Texture Analysis In Predicting Lymph Node Metastasis Of Colorectal Adenocarcinoma

objectives:To explore the value of ¹⁸F-FDG PET/CT texture analysis in predicting lymph node metastasis of colorectal adenocarcinoma,and evaluate the diagnostic efficacy of characteristic texture parameters for lymph node metastasis.Methods:82 patients with colorectal adenocarcinoma underwent ¹⁸F-FDG PET/CT in the Second Affiliated Hospital of Dalian Medical University were retrospectively collected in the study from July 2017 to April 2021,containing 49 cases with lymph node metastasis,33 cases without lymph node metastasis.All cases were confirmed by pathology.Two senior nuclear medicine physicians who didn’t know the pathological results independently analyzed the PET/CT images,selected the largest layer of the lesion from the fusion image,and determined the boundary of the lesion at the same time.In case there were different opinions,negotiated for consistence.All parameters of PET/CT images of patients were standardized,and finally imported the image into the Ma Zda software in BMP format.Then the regions of interest were manually delineated,texture analysis was performed,and relevant parameters were extracted.The feature reduction algorithms including Fishers coefficient,mutual information（MI）,classification error probability integrated average correlation coefficients（POE+ACC）,and combination of the three algorithms（MPF）were performed to select the best discriminating texture feature parameters.The discriminating analysis was executed by principal component analysis（PCA）,linear or nonlinear discriminant analysis（LDA/NDA）and raw data analysis（RDA）to gain misclassification rate to evaluate the accuracy of texture parameter.Difference of the texture parameters in the two groups was verified by sample t test.The diagnostic efficacy was verified by ROC curve.Results:（1）The texture features that were chosen by MPF had the lowest misclassification rate（4.88%～23.17%）for differentiating lymph node metastasis of colorectal adenocarcinom,which was significantly lower than those by Fisher（12.2%～36.59%）、POE+ACC（19.51%～37.71%）and MI（9.76%～37.80%）.（2）The misclassification rate of these four methods were 15.85%～35.37%for RDA,18.29%～36.59%for PCA,10.98%～37.80%for LDA,and 4.88%～19.51%for NDA,the NDA methods had the lowest misclassification rate for differentiating lymph node metastasis of colorectal adenocarcinom.（3）The parameters of Horzl_Shrt REmp,45dgr_Fraction,S（1,1）Dif Varnc,S（1,0）Dif Entrp,S（2,0）Dif Entrp,S（1,1）Dif Entrp,S（4,4）Dif Varnc,Wav En HL_s-1,S（3,3）Dif Varnc,S（2,2）Dif Entrp,Wav En HL_s-3 and Wav En LL_s-1 were higher in the metastasis group than in the group without lymph node metastasis.while the parameters of S（1,1）Correlat,45dgr_Lng REmph,S（1,1）Inv Df Mom,Wav En LL_s-2,Horzl_Lng REmph and Wav En LL_s-1 were lower than the group without lymph node metastasis.（4）The AUC were between 0.623 and 0.750 for the 17 texture parameters,including Horzl_Shrt REmp,45dgr_Fraction,S（1,1）Dif Varnc,S（1,1）Correlat,45dgr_Lng REmph,S（1,1）Inv Df Mom,Wav En HL_s-3,Wav En LL_s-2,S（1,0）Dif Entrp,S（2,0）Dif Entrp,S（1,1）Dif Entrp,Horzl_Lng REmph,S（4,4）Dif Varnc,Wav En HL_s-1,S（3,3）Dif Varnc,S（2,2）Dif Entrp and Wav En LL_s-1.The sensitivity were between 26.53%and 93.88%,the specificity were between 36.36%and 100%.The AUC of S（1,1）Dif Entrp was the largest（0.750）,and the corresponding sensitivity was 87.76%,and the specificity was 57.58%.S（3,3）Dif Varn,S（2,0）Dif Entrp and 45dgr_Fraction had the highest sensitivity（93.88%）for the differentiation of lymphnode metastasis of colorectal adenocarcinoma,and the corresponding AUC was 0.711,0.689 and 0.703,respectively.The specificities was 45.45%,36.36%and 42.42%respectively.The specificities of Wav En HL_s-3 was the highest（100%）,the corresponding AUC was 0.666,and the sensitivity was 26.53%.Conclusions:（1）PET/CT texture analysis could be a new auxiliary method to predict lymph node metastasis of colorectal adenocarcinoma.（2）The texture parameters that can be used to distinguish lymph node metastasis of colorectal adenocarcinoma include Horzl_Shrt REmp,45dgr_Fraction,S（1,1）Dif Varnc,S（1,1）Correlat,45dgr_Lng REmph,S（1,1）Inv Df Mom,Wav En HL_s-3,Wav En LL_s-2,S（1,0）Dif Entrp,S（2,0）Dif Entrp,S（1,1）Dif Entrp,Horzl_Lng REmph,S（4,4）Dif Varnc,Wav En HL_s-1,S（3,3）Dif Varnc,S（2,2）Dif Entrp,Wav En LL_s-1.