¹⁸F-FDG PET dissemination features are prognostic of outcome in diffuse large B-cell lymphoma patients

Objectives: The total metabolic tumor volume (MTV) calculated from baseline ¹⁸F-FDG PET images has been shown to be an early prognostic factor in lymphoma. Yet, patients with similar MTV might present with very different spread of the disease. We introduce new radiomic features characterizing the lesion dissemination in Diffuse Large B-Cell Lymphoma (DLCBL) patients and investigate the prognostic value of such features calculated at baseline.

Methods: Patients from 18- to 59-year old with an advanced stage of DLCBL (Ann Arbor stage of 3 or 4) and for whom ¹⁸F-FDG PET/CT images were available for review were selected from the LNH073B trial. Image processing was performed using the LIFEx software (Nioche et al, Cancer Res 2018). Hyper metabolic regions (ROI) > 1 mL were first automatically detected using a SUV threshold of 3. For each resulting ROI, a threshold of 41% of the ROI SUV_max was used to refine the segmentation. ROI with physiological uptake such as brain and bladder were removed and small nodes could also be manually added using "one click region growing" operations. From that whole-body segmentation, 19 radiomic features (RF) pertaining either to each malignant focus (9 RF) or to the whole body ROI distribution (10 RF) were automatically calculated. These included, among others, MTV, the number of lesions, the distance between the 2 lesions that were the furthest apart (Dmax_patient), the distance between the biggest lesion and the furthest lesion from that bulk (Dmax_bulk), the sum of the distances of the bulky lesion from all other lesions (SPREAD_bulk) and the largest value, over all lesions, of the sum of the distances from a lesion to all the others (SPREAD_patient). The prognostic values of the RF were investigated using ROC to identify the cut-off values, Kaplan Meier (KM) analysis, and Cox regression.

Results: A total of 101 patients were enrolled, half of them treated with R-CHOP and half with R-ACVBP, with no significant impact on outcome. Median MTV, SPREAD_patient, SPREAD_bulk, Dmax_patient, Dmax_bulk were 369 mL, 367 cm, 205 cm, 45 cm, 32 cm respectively. With a median follow-up of 44 months, the 4 year-progression free survival (PFS) and overall survival (OS) were 77% and 85%. Large MTV, Dmax_patient, SPREAD_patient and SPREAD_bulk were adverse factors for PFS (p<0.005, log-rank tests on KM curves). For OS, only MTV (p<0.001) and Dmaxpatient (p<0.01) had prognostic values. In Cox regression, MTV and Dmax_patient remained significant for PFS (p<0.05 and p<0.01) and OS (p<0.01 and p<0.05). Combining MTV (>318 ml) and Dmaxpatient (>58 cm) yielded 3 risk groups for PFS (log-rank test p<0.0001) and OS (p<0.05): high risk with 2 adverse factors (4 year-PFS and OS of 50% and 66%, n=24), intermediate risk with 1 adverse factor (78% and 88%, n=43), and low risk with no adverse factor (96% and 96%, n=28). In R-CHOP arm, MTV and Dmax_patient remained strong prognostic factors whereas in R-ACVBP arm, none of them were significant.

Conclusions: Combining MTV with an RF reflecting the tumor dissemination further improves DLBCL patient risk stratification at staging. Interestingly, the significant prognostic impact of these parameters disappeared in the R-ACBP arm.