A novel radiomic model may be superior to existing tools for predicting outcomes in patients with diffuse large B-cell lymphoma (DLBCL), according to a study recently published in Academic Radiology.

“The decision curve analyses demonstrated that the combined model was superior to the other models across the range of reasonable threshold probabilities, demonstrating the clinical utilities of the constructed models,” the authors wrote.

This retrospective study included 271 patients previously diagnosed with DLBCL. Most (n=152) were male. Mean age was 51 years.

Read more about DLBCL prognosis

Almost all participants (95.6%) received 6 cycles of chemotherapy, while the remaining received 8. After a median follow-up of 34.5 months, 21.4% of the patients either presented with disease progression or relapsed, and 12.5% had died.

The best-combined model that aims to predict the 2-year progression-free survival as the primary endpoint took into consideration data normalization assessed by Z-score, feature selection by analysis of variance (ANOVA), and classification for linear regression by least absolute shrinkage and selection operator (LASSO), as well as 4 different features such as pathology, lactic dehydrogenase levels, Ann Arbor stage, and extranodal involvement.

This tool achieved an area under the curve (AUC) of 0.720, 0.688, and 0.695 in the training, cross-validation, and test dataset phases, respectively. These results are consistently better than those obtained by the international prognostic index (IPI), with values for the AUC of 0.684, 0.684, and 0.584 for training, cross-validation, and test dataset phases, respectively.

This model could provide benefits when identifying high-risk patients for disease progression or relapse after chemotherapy. Its ability to analyze clinical and baseline as well as end-of-treatment positron emission tomography radiomic features leads to superior accuracy over the traditional IPI or Deauville score systems.

“Collecting features from different treatment processes could enrich information and become a novel approach for model construction,” the study concluded.


Cui Y, Jiang Y, Deng X, et al. 18F-FDG PET-based combined baseline and end-of-treatment radiomics model improves the prognosis prediction in diffuse large B cell lymphoma after first-line therapy. Acad Radiol. Published online November 25, 2022. doi:10.1016/j.acra.2022.10.011