| Background and objectPrimary central nervous system lymphoma(PCNSL)is a rare but aggressive nonHodgkin lymphoma(NHL)confined to central nervous system(CNS),accounting for 4～6%all extra-nodal lymphomas.The clinical manifestations and traditional imaging characteristics of PCNSL are similar to those of the other intracranial malignancies,such as high-grade glioma(HGG).Nevertheless,the therapies and prognosis for PCNSL and HGG are quite different.Therefore,it is crucial to establish an accurate diagnosis before treatment.The diagnosis of PCNSL is mainly made through biopsy,which is an invasive and difficult procedure that can cause irreversible damage to brain tissue.In recent years,in addition to conventional morphological imaging(CT and MRI),modern metabolic imaging has been increasingly used to diagnose PCNSL.18Ffluorodeoxyglucose(18F-FDG)PET/CT is one of the most commonly used noninvasive metabolic imaging methods for diagnosing PCNSL,but studies on using 18F-FDG PET/CT imaging features to differentiate between PCNSL and HGG have not been reported.Therefore,in the first part of this study,the imaging features of 18F-FDG PET/CT in PCNSL and HGG patients confirmed by pathology were retrospectively reviewed,and the differential diagnostic value of the 18F-FDG PET/CT radiomics features for distinguishing between the two conditions were explored.Materials and methodsThis study retrospectively included 38 patients with primary central nervous system lymphoma(PCNSL)and 28 patients with high-grade glioma(HGG)confirmed by pathology at Qilu Hospital of Shandong University and Shandong Cancer Hospital.No patient had undergone radiation therapy,chemotherapy,or immunotherapy before 18F-FDG PET/CT imaging.Clinical and pathological data,including gender,age,and pathological type,were collected for all patients.The 18F-FDG PET/CT image data of all patients were imported into LIFEx software in DICOM format.The images were analyzed layer by layer.The lesions were determined by combining the abnormal focal tracer uptake on 18F-FDG PET/CT fusion images and the abnormal density on the CT images.Thereafter,the region of interest(ROI)was delineated manually,and the radiomics features were extracted.Patients from Qilu Hospital of Shandong University and Shandong Cancer Hospital were divided into the training cohort(60.6%,40 cases)and the validation cohort(39.4%,26 cases).The radiomics features extracted were selected by the Least Absolute Shrinkage and Selection Operator(LASSO)model.Radiomics scores were built to differentiate PCNSL and HGG in the training cohort,and the diagnostic ability of the model was validated in the validation cohort.The diagnostic ability of the model was evaluated by the Receiver Operating Characteristic(ROC)curve.ResultsThis study enrolled 66 patients.38 patients had PCNSL,of which 37 had been pathologically diagnosed of B-cell lymphoma and 1 of T-cell lymphoma.28 patients had HGG that pathologically graded as WHO Ⅲ～Ⅳ Based on 18F-FDG PET/CT images,a total of 788 radiomics features were extracted and 8 features were selected by LASSO method.The AUC of the LASSO identification model was 0.962(95%Confidence interval:0.907～1)in training cohort,with a sensitivity and specificity of 95.20%and 88.89%respectively,and an AUC of 0.913(95%CI:0.794-1)with a sensitivity of 87.50%and a specificity of 80.00%in the validation group.ConclusionsThe constructed LASSO regression model based on 18F-FDG PET/CT imaging features can non-invasively distinguish PCNSL from HGG with high diagnostic accuracy.Background and objectDiffuse large B-cell lymphoma(DLBCL)is the most common type of NHL,accounting for about 30～40%of newly diagnosed NHL patients,with various histological subtypes.The application of R-CHOP treatment has significantly improved the survival rate of patients,but some patients still have a risk of relapse or present refractory lymphoma.It is significant to accurately identify DLBCL with poor prognosis and timely change treatment regimen to improve patient prognosis.With the widespread use of 18F-FDG PET/CT in the diagnosis,staging,therapeutic assessment,prognosis prediction,and relapse monitoring of system lymphoma,establishing a predictive model based on this imaging modality is currently a research hotspot.Therefore,this study retrospectively analyzed the baseline texture features of 18F-FDG PET/CT in DLBCL patients confirmed by pathology,and attempted to establish a predictive model based on 18F-FDG PET/CT imaging features,in order to accurately predict patient prognosis,change treatment regimens in a timely manner,and improve patient outcomes.Materials and methods106 patients with confirmed diffuse large B-cell lymphoma(DLBCL)at Qilu Hospital of Shandong University were included in the study.All patients had not received any treatment before baseline 18F-FDG PET/CT examination.Clinical and pathological data were collected for all patients.Patients were divided into the progression group(2-y PFS 1)and the non-progression group(2-y PFS 0)with the 2-y progression-free survival(PFS)as the endpoint.The baseline 18F-FDG PET/CT images in DICOM format were imported to the LIFEx software,and the lymphoma lesions were delineated manually.Radiomics features were extracted,and metabolic parameters of 18F-FDG PET/CT imaging were automatically calculated by the LIFEx software.The Synthetic minority oversampling technique(SOMTE)was used to handle the imbalance between the progression group and the non-progression group.The Monte Carlo(MC)cross-validation method with a training-to-validation set ratio of 80:20 was used to trained the model,and the final prediction model was established using three-layer machine learning(ML).The corrected independent samples t-test or the Mann-Whitney U test was used for statistical analysis of the metabolic parameters of the two groups.The predictive performance of the ML model and various metabolic parameters was evaluated using the ROC curve.ResultsThis study included a total of 106 patients(51.34±14.84 years old),the total number of drawn lesions was 1175,and the highest number of lesions in a single patient was 118.78 patients(73.58%)had no progression and survived within 2 years.The dataset was composed of 156 samples after SMOTE processing,with 50%(78 cases)of the samples in each of the progression and no progression groups.The prediction model obtained through three-layer ML had an AUC of 0.819(95%CI:79.43～82.58),a sensitivity of 80.26%and a specificity of 79.86%in predicting 2-year PFS of DLBCL.The AUC of SUVmax,SUVmean,SUVpeak,MTV and TLG in predicting PFS in DLBCL patients within 2 years were 0.435(95%CI:0.345-0.525),0.468(95%CI:0.377-0.560),0.438(95%CI:0.348-0.529),0.608(95%CI:0.519-0.697)and 0.585(95%CI:0.495-0.675),respectively.The sensitivities were 46.1%,48.4%,46.9%,55.3%,and 54.2%,and the specificities were 47.3%,48.4%,46.9%,55.4%,and 54.2%respectively.ConclusionsThe ML model based on 18F-FDG PET/CT radiomics features has higher predictive efficacy for 2y-PFS of DLBCL patients compared to traditional PET/CT metabolic parameters. |
PDF Full Text Request