Objectives: The current study sought to determine the potential use of the CT radiomics model in predicting overall survival in DLBCL patients.
Methods: The CT images and clinical data of DLBCL patients receiving chemotherapy from January 2013 to May 2018 were retrospectively analyzed, and 130 patients were included and categorized as training cohort (n=91) and validation cohort (n=39) at a 7:3 ratio. The CT radiomics features were extracted, and the Rad-score was calculated using the LASSO (least absolute shrinkage and selection operator) algorithm. Univariate and multivariate Cox regression was used to screen independent risk factors, and then a nomogram model was developed jointly with the Rad-score. The ROC(operating characteristic curve), calibration curve, and decision curve assessments were utilized to assess the model's effectiveness, accuracy, and clinical significance in predicting OS.
Results: In total, 878 CT radiomics features were obtained from each patient, and 15 features highly correlated with OS in DLBCL patients were screened to calculate the Rad-score used to predict OS. Patients with Rad-score <-0.51 had a shorter overall survival time, and those with Rad-score >-0.51 had a longer overall survival time. A nomogram model was constructed by combining independent risk factors (Ann Arbor staging, IPI score, PS, effectiveness) based on multivariate analysis and Rad-score. In the training and validation cohorts, the AUC values of the nomogram model for predicting 3 and 5 years OS were 0.860 and 0.810, respectively, 0.838 and 0.816 which were higher than the Rad-score (0.744 and 0.763, respectively, 0.787 and 0.563). Furthermore, the calibration and decision curve evaluations revealed that the nomogram model strongly agrees and has a high clinical value in predicting OS in DLBCL patients.
Conclusion: The nomogram model based on clinical characteristics and CT radiomics features have a better prediction efficacy for overall survival following first-line treatment in DLBCL patients, and it exceeds the Rad-score model.