Healthy Organs Uptake on Baseline 18F-FDG PET/CT as an Alternative to Total Metabolic Tumor Volume to Predict Event-Free Survival in Classical Hodgkin's Lymphoma

Purpose Healthy organs uptake, including cerebellar and liver SUVs have been reported to be inversely correlated to total metabolic tumor volume (TMTV), a controversial predictor of event-free survival (EFS) in classical Hodgkin's Lymphoma (cHL). The objective of this study was to estimate TMTV by using healthy organs SUV measurements and assess the performance of this new index (UF, Uptake Formula) to predict EFS in cHL. Methods Patients with cHL were retrospectively included. SUV values and TMTV derived from baseline 18F-FDG PET/CT were harmonized using ComBat algorithm across PET/CT systems. UF was estimated using ANOVA analysis. Optimal thresholds of TMTV and UF were calculated and tested using Cox models. Results 163 patients were included. Optimal UF model of TMTV included age, lymphoma maximum SUVmax, hepatic SUVmean and cerebellar SUVmax (R2 14.0% - p < 0.001). UF > 236.8 was a significant predictor of EFS (HR: 2.458 [1.201–5.030], p = 0.01) and was not significantly different from TMTV > 271.0 (HR: 2.761 [1.183–5.140], p = 0.001). UF > 236.8 remained significant in a bivariate model including IPS score (p = 0.02) and determined two populations with different EFS (63.7 vs. 84.9%, p = 0.01). Conclusion The Uptake Formula, a new index including healthy organ SUV values, shows similar performance to TMTV in predicting EFS in Hodgkin's Lymphoma. Validation cohorts will be needed to confirm this new prognostic parameter.


INTRODUCTION
Hodgkin's Lymphoma (HL) affects young adults and represents about 2.3 cases per 100,000 people per year, with an associated mortality of 0.4 cases per 100,000 per year (1). Despite treatment, about 20% of HL patients still relapse (1). 18 F-Fluorodeoxyglucose ( 18 F-FDG) Positron Emission Tomography (PET) coupled with Computed Tomography (CT) plays a central role in HL patients management, whether in staging or response assessment settings (1).
PET-derived parameters, volumetric ones above all, have been proposed to refine prognosis prediction of HL (2). The role of Total Metabolic Tumor Volume (TMTV) is debated in Hodgkin's Lymphoma. It has been reported as a negative prognostic factor in early-stage HL treated with ABVD (Adriamycin, Bleomycin, Vinblastine, Dacarbazine) regimen (3)(4)(5) and HIV-associated HL (6). However, some publications reported no association between TMTV and Progression-Free Survival (PFS) in advanced-stage HL when treated with escalated BEACOPP (7). Furthermore, TMTV threshold varies from one study to another (3-5, 7, 8).
The drawbacks of TMTV calculation (results depending on the segmentation method (9), time required for delineation (10), difficulty in evaluating bone involvement) led to ponder other prognostic markers, such as healthy organ 18 F-FDG uptake. In 2010, Hanaoka et al. (11) reported an inverse correlation between cerebellar uptake and total lesion glycolysis (TLG) in a population with aggressive lymphoma. The mechanism underlying this phenomenon is poorly understood but could correspond to a metabolic theft of 18 F-FDG by the tumor mass. Because TLG is correlated with TMTV, Godard et al. (12) speculated that cerebellar metabolism might have a prognostic value and suggested to normalize cerebellar 18 F-FDG uptake to hepatic 18 F-FDG uptake to account for differences between PET/CT systems. This index has been shown to be a prognostic parameter for PFS prediction in diffuse large-B-cell lymphoma (10) and follicular lymphoma (12). Normalization to liver was not optimal: liver 18 F-FDG uptake was also negatively correlated with TMTV as was cerebellar 18 F-FDG uptake (r = −0.34 and r = −0.42, respectively) (10). These two healthy organs could thus prove useful in predicting prognosis.
The objective of this study was to model TMTV by integrating healthy organ uptake data in classical HL (cHL). The resulting estimate was then tested for EFS prediction.

Study Population
This study was approved by the Ethical Committee of Fondazione Policlinico Universitario A. Gemelli IRCCS (study code: 3834). All included subjects signed an informed consent form. All procedures performed were in accordance with the ethical standards defined by the 1964 Helsinki Declaration and its later amendments.
All patients with HL referred to our Institution for their baseline 18 F-FDG PET/CT between September 2010 and January 2020 were retrospectively screened. Inclusion criteria were as follows: histologically proven cHL; baseline PET/CT performed within 4 weeks prior to treatment. Exclusion criteria were: Recent history of other cancer <1 year; Nodular lymphocytepredominant Hodgkin lymphoma histology [slow growing LH subtype with completely different prognosis (13)]; any factor interfering with measurement of cerebellar uptake, liver uptake or TMTV: cerebellum not fully included in field of view, movement artifacts, extensive surgically resected disease before staging PET/CT, diffuse lymphomatous involvement of liver or brain lymphoma; nonobservance of a fasting period of at least 6h before 18 F-FDG administration; glycemia > 2.0 g/l; no follow-up available after staging PET/CT; nonstandard treatment regimen.
The following clinicobiological data were collected: date of birth, date of diagnosis, date of last observation, HL subtype, Eastern Cooperative Oncology Group (ECOG) performance status, International Prognostic Score (IPS) items (age; sex; Ann Arbor stage; serum albumin levels, white blood cell count, lymphocyte count, hemoglobinemia at baseline), first-line treatment, event-free survival (EFS: time interval between date of diagnosis and the event of progression, recurrence, change of therapy or death) and overall survival (OS: time interval between date of diagnosis and death). Imaging data collected included: date of staging PET/CT, PET/CT system, administered 18 F-FDG activity, glycemia levels.

F-FDG PET/CT Acquisition and Measurements
After verification of patients' blood glucose levels, baseline 18  PET3 corresponded to a Biograph mCT PET/CT scanner (Siemens Healthineers): images were acquired in 2.5 min per bed position and reconstructed on a 400 × 400 matrix using 3D Ordered Subset Expectation Maximization algorithm (3D-OSEM, 2 iterations, 21 subsets, voxel size: 3.2 × 3.2 × 5 mm 3 ) with PSF and TOF. A gaussian filter was also applied (3D isotropic Gaussian kernel of 2 mm full width at half-maximum).
CT acquisition protocol was the same for the three machines: 120 kV, 50 mAs, reconstruction on a 512 × 512 matrix with a voxel size of 0.97 × 0.97 × 3 mm 3 . PET/CT images were acquired at least from skull base to proximal thighs.  PET/CT were displayed on dedicated interpretation consoles (Syngo.via for SUV measurements and version 7.0.5 of MIM Encore Software for volumetric parameters). The following data were collected: (1) cerebellar SUV max , (2) hepatic SUV mean , (3) TMTV, (4) lymphoma maximal SUV max (lesion SUV max , henceforth denoted as L).
After a first visual check using a rainbow 10 point-color scale, enclosing region of interest (ROI) were drawn on areas with highest visual uptake, excluding any voxel of the neighboring brain hemispheres. The highest SUV max of all these ROI corresponded to cerebellar SUV max . A default spherical ROI (2-cm diameter) was positioned on the right liver to measure its SUV mean . The lymphoma maximal SUV max was determined manually by an experienced nuclear medicine physician and was defined as the SUV max of the hottest nodal lesion. TMTV was measured using a PET segmentation tool (LesionID, version 7.0.5 of MIM Encore Software Inc., Cleveland, OH). As previously described (14), the software proceeds in 4 steps: first, a PET Response Criteria in Solid Tumors (PERCIST)-based background threshold (liver) thresholding was applied; second, a VOI encompassing all detected lesions (above the threshold) was automatically drawn. The detected lesions could thus include bone and spleen, depending on their uptake; third, a second thresholding at 41% of the SUVmax of the detected lesions was applied to determine lesions' boundaries; finally, physicians were required to reject false positive lesions before computation of TMTV.

ComBat Harmonization
The "batch effect" introduced by the use of 3 different PET/CT machines was compensated using a validated statistical harmonization method (15) implemented on RStudio (16).
ComBat was applied on log transformed data, followed by exponentiation to improve the algorithm effectiveness (15), and ensure positive values. TMTV, hepatic SUV mean (H), cerebellar SUV max (C) and lesion SUV max , (L) were harmonized. Reference batch was set to PET3. An example is presented in Figure 1.

Statistical Software
If not stated otherwise, the following statistical analyses were performed on Xlstat (2020, Addinsoft, New York, USA). p-value threshold for significancy was set at 0.05.

TMTV Modelization
An ANOVA analysis was used to model the TMTV from the following 5 clinicobiological data: age, blood glucose, H, C, L. The selection of the optimal model was based on the R² value with a number of allowed parameters ranging from 2 to 5. The resulting formula is hereafter referred to as the Uptake Formula (UF). Significance was assessed by F-statistic.

Analysis
Optimal cut-off for TMTV, UF and IPS were calculated using CutoffFinder (17)

RESULTS
Two-hundred and fifty-four patients were retrieved from the database (Figure 2). Among them, 4 had a nodular-lymphocytepredominant HL (1.6%) and were excluded. Among the remaining 250 patients, 77 were excluded due to the impossibility of measuring the needed parameters (cerebellum outside the field of view, surgically resected disease, corrupted data). Seven

Event-Free Survival Analysis
Optimal threshold for UF, TMTV and IPS were 236.8, 271.0, and 2.0, respectively ( Table 3). The three parameters were significant predictors of EFS with HR between 2.050 and 2.761. When pooled with IPS, both UF and TMTV remained significant predictors of EFS (p = 0.022 and p = 0.004, respectively).
EFS survival curves based on UF are presented in Figure 3.

Healthy Organs 18 F-FDG Uptake Values and Derived Formula
No significant differences in hazard ratio were found between UF and TMTV. UF remained significant at bivariate analysis when adding IPS score with an HR of 2.3 (derived EFS of 84.9 vs. 63.7%). The metabolic theft hypothesis for which FDG-avid tumor masses would deprive healthy organs of 18 F-FDG was investigated in two papers on diffuse large-cell B-cell and follicular lymphomas (10,12). Cerebellar and hepatic 18 F-FDG uptake values were reported to be inversely correlated to TMTV. The optimal model we found to estimate TMTV is coherent with these findings: both liver and cerebellum coefficients are negative, meaning an inverse correlation with TMTV estimate. The addition of these two parameters contributed to the significance of the model, which, however, solves only part of the variability in TMTV (R² of 14.0%), explaining the slight difference between TMTV and UF optimal thresholds. Besides healthy organ uptake, age and tumoral 18 F-FDG uptake were also selected in the model. Apart from the study by Angelopoulou et al. (18), who reported that SUV max was predictive of PFS in a study of 162 patients with HL, other studies reported no significance (3). The lack of harmonization is probably one of the overriding factors for those results.
Both SUV mean and SUV max were used. SUV mean , which is less sensitive to noise, was preferred for the measurement of hepatic uptake to promote reproducibility: as the liver is a homogeneous organ, variations in the positioning of the ROI have little impact on SUVmean measurement as already noted in a previously published study (12). The cerebellum has on the other hand a heterogeneous uptake mainly concentrated in the gray matter. The measurement of SUVmean would have required a precise contouring of this structure that could have introduced contouring bias (19). SUVmax, which relies on only a single pixel was then chosen to ensure reproducibility, as already demonstrated in another study (12).
UF thus had the advantage of speed of calculation, requiring only 3 measurements of SUV values vs. several minutes for TMTV [6.2 min on average, ranging from 0.4 to 21.6 min in the study by Ilyas et al. (20)].

Metabolic Tumor Volume as Prognostic Factor
Some studies have already investigated the prognostic value of TMTV in HL (3-7, 18, 21-24) with conflicting results, presumably related to the difference in patients' therapeutic management and the low number of events encountered in HL (25). Most studies reported a significant ability of TMTV to predict PFS (3,5,22), with an overall HR calculated by Frood et al. (25) of 2.13 (CI 95% 1.53-2.96). These results were however associated with high levels of heterogeneity. Segmentation methods and cut-offs varied greatly [TMTV cutoff from 89 ml (22) to 225 ml (21)] and no test-retest of these thresholds were performed. Moreover, the cut-off determination method is another aspect that needs to be addressed. As pointed out by Schöder and Moskowitz (26), most studies rely on Receiver Operating Curves to determine variables' cut-offs, neglecting censored data and leading to inappropriate results (27)(28)(29). To overcome this issue, a survival-based cut-off method was used in this study. Even if our TMTV threshold (271 cm 3 ) was higher than the previously mentioned ones, we found a similar HR to previously reported ones.

Harmonization
Pooling images from different scanners is not simple, as many quantitative biomarkers (SUV, TMTV) are sensitive to a scanner effect (15,30). Although procedures were proposed to harmonize image quality (31), a dedicated reconstruction requiring raw data storage would be needed and would mostly be not feasible in a retrospective setting (15). To counteract this batch effect, the ComBat harmonization method, initially introduced in the field of genomics (32), has been proposed (15) and used (33). ComBat is a datadriven method that does not require phantom acquisitions to estimate the scanner effect but requires data from the different sites with sufficient sample size. It always theoretically improves the alignment of the mean and standard deviation of the distributions, given the criterion optimized by the method (15).
In our study, we chose to harmonize SUV and TMTV values using PET3 scanner as a reference for several reasons: it is the most recent machine among the three, corresponding to currently available technology in PET scanners; furthermore, the majority of patients was scanned on the PET3, so TMTV and SUV values were not modified for the majority of patients. Harmonization allowed us to study cerebellar and hepatic uptake independently, without having to use a ratio for normalization purposes. The use of a ratio disturbed the correlation between healthy organ and TMTV in the article by Morland et al. (10), but was still necessary to ensure good inter-machine agreement. The ComBat harmonization allowed us to overcome this problem.

Limitations
Some limitations of this study can be pointed out. This exploratory retrospective study, lacks an external and/or prospective validation cohort which would be desirable to confirm our findings. Moreover, the cohort is heterogeneous due to different stages at diagnosis that may have interfered with the performance of the parameters tested. Nevertheless, large retrospective studies are commonly designed to evaluate prognostic parameters in lymphoma, needing long-term follow-up to register a significant number of adverse events.

CONCLUSION
The Uptake Formula, a new index including healthy organ uptake values, shows similar performance to TMTV in predicting EFS in Hodgkin's Lymphoma. Validation cohorts will be needed to confirm this new prognostic parameter.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Ethical Committee of Fondazione Policlinico Universitario A. Gemelli IRCCS (Study Code: 3834). The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.

FUNDING
SA and ET were supported by the Italian Ministry of Health (GR-2019-12370372).