Differential Diagnosis Of Lymphoma Involving The Anterior Mediastinum Based On ¹⁸F-FDG PET/CT Radiomics

Objective:For lymphoma involving or primary in the anterior mediastinum,due to its rich pathological types and complex clinical data,it cannot be effectively distinguished only by traditional imaging indicators or existing clinically relevant indicators,so early and correct identification of the anterior mediastinum The pathological type of lymphoma has great value for clinical diagnosis and treatment options.This study started with the radiomics features extracted from the clinical baseline 18F-FDG PET/CT,and analyzed the prediction models of the three types of lymphoma involving the anterior mediastinum through machine learning algorithms,so as to provide certain predictions for the differential diagnosis of anterior mediastinal lymphoma value.Methods:From November 2017 to February 2023,93 patients with lymphoma involving the anterior mediastinum who underwent ¹⁸F-FDG PET/CT baseline examination in our hospital from November 2017 to February 2023 and had clear pathological diagnosis and complete clinical data were retrospectively analyzed.Three groups of patients with lymphoblastic lymphoma（T LBL,n=32）,Hodgkin’s lymphoma（HL,n=25）,and diffuse large B-cell lymphoma（DLBCL,n=23）were included in the analysis and divided into L₁（DLBCL-HL）group,L₂（DLBCL-TLBL）group and L₃（TLBL-HL）group,three groups were analyzed for differential diagnosis of two classifications.The study patients were randomly divided into groups,70%of the patients in the training group,and the remaining 30%were used as the test group.Features are selected using m RMR and LASSO.First,m RMR is performed on redundant and irrelevant features,and 30 features are retained.Then LASSO is used,and the bestλvalue is selected through cross-validation,and the optimized feature subset is selected to build the final model.To compare the diagnostic performance of the established Radiomics scores（Radiomics scores,Rad-score）model.Receiver operating curve（ROC）and area under the curve（AUC）were used to evaluate the differential diagnosis performance between the models.Results:Afterfeaturescreening,whenλ_L1=0.0336055419891073,λ_L2=0.0246884888278534 andλ_L3=0.0175238647432615,eight imaging features were extracted from the three groups of models to construct the Rad-score model.In the differential diagnosis of DLBCL and HL,when the critical value of Rad-score was0.030 and 0.019,the values of AUC in training set and test set were 0.72 and 0.69respectively.In the differential diagnosis of DLBCL and T-LBL,when the critical value of Rad-score was taken-0.031 and-0.505,the values of AUC in training set and test set were 0.83and 0.81respectively.In the differential diagnosis of T-LBL and HL,when the critical value of Rad-score was 0.342 and 0.787,the values of AUC in training set and test set were 0.89 and 0.88 respectively.In the differential diagnosis of lymphoma among the three groups,the difference of Rad-score between the two groups was statistically significant（P<0.05）.Conclusion:Based on the omics parameters extracted by 18F-FDG PET/CT,the three-group identification model can distinguish between diffuse large B-cell lymphoma,Hodgkin lymphoma and T-lymphoblastic lymphoma involving the anterior mediastinum significance.In the differential group of DLBCL-HL,DLBCL-TLBL and TLBL-HL,log.sigma.3.0.mm.3D＿glcm＿Inverse Variance,log.sigma.2.0.mm.3D＿glrlm＿Long Run Emphasis and wavelet.LLL＿ngtdm＿Strength has higher differential value.