A Model For Differentiating Primary Intestinal Lymphoma From Crohn’s Disease Based On ¹⁸F-FDG PET/CT:Establishment And Validation

BachgroundPrimary intestinal lymphoma(PIL)is a rare primary malignancy originated from intestine.For it has the varing clinical symptoms and signs,and most of them are lack of specificity,it is very difficult to make the accurate diagnosis,and it needs to be differentiated from a series intestinal diseases.Clinically the misdiagnosis rate of PIL is high,and a considerable portion of them is misdiagnosed as Crohn’s disease(CD).The advantages of 2-Fluorine-18-Fluoro-2-deoxy-D-glucose(18F-FDG)positron emission computed tomography/computedtomography(PET/CT)in diagnosis,as well as the assessment of lymphoma have gained accepted globally,besides,in the lasting decade,the value of 18F-FDG PET/CT in diagnosis,extent of disease,as well as the assessment of curative effect of Crohn’s disease gradually has been coming in notice.Although 18F-FDG PET/CT has distinct advantages in diagnosis and assessment of PIL and CD respectively,to our best knowledge,there is no published literature about using 18F-FDG PET/CT in differentiating PIL from CD.Now we try to establish a mathematical model which is mainly based of the 18F-FDG PET/CT characters in order to differ PIL from CD,and subsequently evaluate the effectacy of this model.Part Ⅰ Establishment of A model for differentiating primary intestinal lymphoma from Crohn’s disease based on 18F-FDG PET/CTObjectiveTo establish a mathematical model based on 18F-FDG PET/CT data,and assess its potency for differentiating primary intestinal lymphoma(PIL)from Crohn’s disease(CD).Materials and Methods22 patients with PIL and 67 with CD undergone 18F-FDG PET/CT in our center from January 2005 to August 2016 were included in our reserch,the clinical and imagilogy data of the patients were analysed retrospectively.The patients clinical date(gender and age),lesions distribution(quantity and location),18F-FDG PET/CT characters(SUVmax of intestinal lesion,bone marrow,right lobe of liver and regional lymph node respectively),corresponding CT characters(ring-like intestinal wall thickening,localized mass,segmental bowel dilatation,identifiable mucosa folds thickening,bowel stiffness,mesenteric change around the intestinal lesion,diameter of the regional lymph node)and complication(bowel obstruction,perforation or fistula,abscite,perianal lesion)were colleted to establish 3 individual mathematical models(clinical data and 18F-FDG PET characters,clinical data and CT characters,comprehensive respectively)by using single factor analysis and logistic regression.Subsequently the efficacy of the above 3 models were analysed and compared by using receiver operating characteristic(ROC)curve to select the best one,then the optimal predicting diagnostic value was explored by using Yorden’s index method.Result1.Single factor analysis of the indices of the 2 groups 18 indices of the 2 groups were studied by using single factor analysis.Among them,only 10 variables were demonstrated to have statistically significant difference,as below:age(AGE,t=4.191,P<0.001)the SUVmax ratio of the intestinal lesion to the right lobe of liver(R1,t=-3.864,P<0.001),localized mass(CT2,X2=4.469,P=0.035),segmental bowel dilatation(CT3,X2=7.089,P=0.008),identifiable mucosa folds thickening(CT4,X2=16.838,P<0.001),bowel stiffness(CT5,X2=4.559,P=0.033),mesenteric change around the intestinal lesion(CT6,X2=5.268,P=0.022),the SUVmax of the abdominal regional lymph node(LN1,x2= 10.610,P=0.005),the maximum diameter of the abdominal regional lymph node(LN2,X2=14.092,P=0.001),and perianal lesion(CPX4,X2=4.385,P=0.036).2.Logistic regression of the selected indices and establishment of the mathematical model The above 10 indices were operated by 3 different purposes by using logistic regression.In model A(clinical data and 18F-FDG PET characters),only 2 indices(AGE and R1)were found to have statistically significant difference.In model B(clinical data and CT characters),4 indices(AGE,CT4,CT6,CPX4)were found to have statistically significant difference.In model C(comprehensive),4 indices(AGE,R1,CT4,CT6)demonstrated to have statistically significant difference.Therefrom 3 individual mathematical models were respectively established:P=1/(1+e-Z),where ZA=-5.562+0.070×AGE+0.304×R1,ZB=-1.886+0.067×AGE-2.544×CT4-1.877×CT6+1.751×CPX4,Zc=-3.289+0.064xAGE-1.660xCT4-1.608xCT6+0.273xR1 respectively.3.The comparison of the 3 mathematical modelsThe above 3 models were analysed by using ROC curve,thearea under the curve(AUC)of model A,B and C was 0.799,0.809 and 0.919 respectively.And the effectacy of model C was significantly better than model A and model B(z=2.501 and 1.878 respectively;P=0.012 and 0.043 respectively),there was no statistically difference between model A and model B in effectacy(z=0.116,P=0.901).4.Determination of the optimal predicting diagnostic value of the model Using a optimal predictive diagnostic value(P)of 0.277 determined by Yorden’s index method for differentiating.When P>0.277 diagnosed with PIL,and P<0.277 with CD.ConclusionA mathematical model for differentiating PIL from CD mainly based on 18F-FDG PET/CT characters is successfully established.The result showed the differentiating effectacy of this model was superior to that mainly based on 18F-FDG PET/CT or CT one respectively.That indicated the comprehensive model has a potential value for clinical application.Part II Validation and comparison of the new differentiating diagnostic model based on 18F-FDG PET/CT and the existing modelObjectiveTo validate the new differentiating diagnostic model based on 18F-FDG PET/CT which was established in the Part I and the existing model(Zhang’s model),and to compare their potential effectacy in differentiating PIL from CD.Materials and Methods1.The clinical information of the patients in the validating groupAll the patients in the modeling group were confirmed histopathologically by surgery or biopsy.Among them,6 cases were PIL(3 were EATL,2 were DLBCL,and the left one was MALToma),and 14 cases were CD(10 were in active stage and 2 were in remission stage respectively).2.The differentiating effectacy of Zhang’s modelTheROC-AUC of Zhang’s model was 70.8%(95%CI:0.466～0.887).As 0.5 was defined as the cutoff value for differentiating PIL from CD,the sensitivity and the specificity of the model was 66.67%(2/3)and 76.47%(13/17)respectively,besides,the positive predicting value and the negative predicting value was 33.33%(2/6)and92.86%(13/14)respectively.3.Validation of the new modelValidated the new model by using the data of 20 patients of the validating group,calculated the probability of PIL and drew the ROC curve with the cutoff value of 0.277.The sensitivity and the specificity of the model was 60.00%(3/5)and 80.00%(12/15)respectively,besides,the positive predicting value and the negative predicting value was 50.00%(3/6)and 85.71%(12/14)respectively.4.Comparison of the new model and the Zhang’s model Analysed and compared the ROC of the new model and the Zhang’s model.Resulting the ROC-AUC of the Zhang’s model was 70.8%(95%CI 为 0.466～0.887)and the new model was 83.3%(95%CI 为 0.601～0.960)separately.There was no significant difference between these 2 models in differentiating effectacy(z=0.748,P=0.4545).Whencomparing the 2 models by using 0.5 as the cutoff value of the Zhang’s model and P=0.277 as of the new model,there were no significant differencies both in sensitivity and specificity(66.67%vs.60.00%,X2=0.343,P=0.558;76.47%vs.80.00%,X2=0.373,P=0.541;respectively).