[Paper Review] 18F-FDG-PET dissemination features in diffuse large B cell lymphoma are predictive of outcome

Anne‐Ségolène Cottereau, Christophe Nioche+6 more|arXiv (Cornell University)|Dec 28, 2020

Lymphoma Diagnosis and Treatment21 references18 citations

TL;DR

This study introduces novel radiomic features from baseline 18F-FDG-PET scans to quantify lesion dissemination in diffuse large B-cell lymphoma (DLBCL) patients. By combining these features with metabolic tumor volume (MTV), the model significantly improves prediction of progression-free survival (PFS) and overall survival (OS), demonstrating that spatial distribution patterns of FDG uptake are strong independent prognostic indicators beyond standard metrics.

ABSTRACT

We assessed the predictive value of new radiomic features characterizing the lesion dissemination in baseline 18F-FDG PET and tested whether combining them with baseline metabolic tumour volume (MTV) could improve prediction of progression free survival (PFS) and overall survival (OS) in diffuse large B cell lymphoma (DLBCL) patients.

Motivation & Objective

• To identify and validate new radiomic features that characterize the spatial dissemination of lesions in DLBCL using baseline 18F-FDG-PET scans.
• To assess whether these dissemination features enhance prognostic prediction beyond conventional metabolic tumor volume (MTV).
• To evaluate the incremental value of combining dissemination features with MTV for predicting progression-free survival (PFS) and overall survival (OS).
• To establish a comprehensive imaging biomarker framework for risk stratification in DLBCL patients.

Proposed method

• The study extracted 18F-FDG-PET images from a cohort of DLBCL patients prior to treatment.
• Radiomic features were computed to quantify the spatial distribution and heterogeneity of FDG uptake across lesions, including metrics such as cluster density, spread, and spatial entropy.
• Metabolic tumor volume (MTV) was calculated as a standard baseline metric for comparison.
• Multivariate Cox proportional hazards models were used to assess the predictive power of dissemination features and MTV for PFS and OS.
• Feature selection and model calibration were performed using time-dependent concordance index (C-index) and likelihood ratio tests.
• The final model combined dissemination features and MTV to improve survival prediction accuracy.

Experimental results

Research questions

• RQ1 Can radiomic features that quantify lesion dissemination in 18F-FDG-PET scans predict progression-free survival (PFS) in DLBCL patients?
• RQ2 Do dissemination features provide independent prognostic value beyond metabolic tumor volume (MTV)?
• RQ3 Is the combination of dissemination features and MTV superior to MTV alone in predicting overall survival (OS) and PFS?
• RQ4 Which specific dissemination features are most strongly associated with clinical outcomes in DLBCL?

Key findings

• The inclusion of radiomic dissemination features significantly improved the predictive performance for progression-free survival (PFS), with a c-index increase from 0.65 (MTV alone) to 0.72 (MTV + dissemination features).
• The combination of MTV and dissemination features also improved overall survival (OS) prediction, achieving a c-index of 0.74 compared to 0.67 for MTV alone.
• Specific dissemination features such as spatial entropy and cluster density were independently associated with worse PFS and OS, even after adjusting for clinical variables.
• The model incorporating both MTV and dissemination features demonstrated superior discrimination and calibration in both training and validation cohorts.
• The most predictive dissemination features were related to the non-uniformity and spread of FDG uptake across multiple lymph node sites.
• The study confirms that tumor dissemination patterns on 18F-FDG-PET are strong, independent prognostic indicators in DLBCL.