Tumor metabolic heterogeneity based on ¹⁸F-FDG PET/CT is a predictor of outcome in diffuse large B-cell lymphoma

Abstract

Background

Tumor metabolic heterogeneity (MH) assessed by 18-fluorine fluorodeoxyglucose positron emission tomography computed tomography (¹⁸F-FDG PET/CT) is established as a marker of drug resistance in solid tumors. Its prognostic significance in diffuse large B-cell lymphoma (DLBCL) remains inadequately explored. This study aims to investigate the prognostic value of tumor MH assessed by ¹⁸F-FDG PET/CT in DLBCL.

Methods

This study retrospectively included 297 patients diagnosed with DLBCL at the Third Affiliated Hospital of Soochow University from August 2012 to December 2022. We evaluated tumor MH using the area under the curve of cumulative standardized uptake value-volume histogram (AUC-CSH). In addition to obtaining conventional PET metabolic metrics, including maximum standardized uptake value (SUVmax), mean standardized uptake value (SUVmean), total metabolic tumor volume (TMTV), and total lesion glycolysis (TLG), we also collected baseline clinical data. This study developed a survival prediction model based on Cox regression analysis and compared its prognostic performance with the National Comprehensive Cancer Network-International prognostic index (NCCN-IPI). We evaluated the model’s calibration, discrimination, and clinical utility, and conducted an internal validation.

Results

Multivariable Cox regression analysis indicated age (hazard ratio [HR], HR,1.62, 95% confidence interval [CI], 1.13 ~ 2.34, P = 0.01; HR, 2.15, 95% CI, 1.35 ~ 3.42, P = 0.001), TMTV (HR, 3.24, 95% CI, 1.56 ~ 6.73, P = 0.002; HR, 3.35, 95% CI, 1.30 ~ 8.62, P = 0.012), and AUC-CSH (HR, 0.64, 95% CI, 0.45 ~ 0.92, P = 0.015; HR, 0.59, 95% CI, 0.38 ~ 0.92, P = 0.018) as independent predictive factors for both progression-free survival (PFS) and overall survival (OS). Lower AUC-CSH value corresponded to higher tumor MH, indicating poorer survival outcome. The combined model outperformed the NCCN-IPI, achieving C-index of 0.688, 0.630 for PFS and 0.721, 0.688 for OS. The calibration curve showed good consistency, and the decision curve analysis (DCA) also confirmed its potential application value in individualized treatment decisions.

Conclusions

Tumor MH, assessed by ¹⁸F-FDG PET/CT, demonstrates potential as a prognostic biomarker in DLBCL. Its incorporation into predictive models may improve individualized prognostic evaluation and facilitate more precise treatment stratification for these patients.

Trial registration

This research has been retrospectively registered in the Ethics Committee institutional of the Third Affiliated Hospital of Soochow University [approval number (2022) KD155].

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-025-15149-x.

Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most prevalent pathological subtype of aggressive non-Hodgkin’s lymphoma and accounts for the largest share of lymphoma-related deaths worldwide [1, 2].The standard first-line chemoimmunotherapy of rituximab combined with cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) has significantly improved patient prognosis, achieving long-term disease control in nearly 90% of patients with limited-stage disease and up to 60% of patients with advanced-stage disease [3]. However, patients who fail R-CHOP treatment typically have poor outcomes, with a median overall survival (OS) of only 6 months [4]. Furthermore, treatment options for patients with relapsed or refractory (R/R) disease are limited. While high-dose chemotherapy combined with autologous stem cell transplantation can provide the best curative opportunity for this population, their prognosis remains discouraging, with only a minority achieving remission through salvage therapy [5, 6]. Patients who did not experience R/R within two years after initial treatment have a survival rate similar to that of the general population. This underscores the critical importance of timely and accurate identification of R/R patients before treatment, as well as the optimization of treatment regimens [7]. Currently, widely used models based on clinical data, such as the International Prognostic Index (IPI) and the National Comprehensive Cancer Network (NCCN)-IPI, are insufficient in effectively identifying high-risk subsets with poor outcomes to chemoimmunotherapy.

18-fluorine fluorodeoxyglucose positron emission tomography computed tomography (¹⁸F-FDG PET/CT) has become an increasingly irreplaceable tool for the assessment of disease burden, staging, and treatment response in DLBCL [8, 9]. The prognostic significance of quantitative PET metrics, including metabolic tumor volume (MTV) and total lesion glycolysis (TLG), in DLBCL has been substantiated by previous validation studies [10–12]. Indeed, relying solely on PET volumetric parameters such as MTV and TLG−which reflect tumor burden−may be insufficient for predicting prognosis in DLBCL. This is because these parameters fail to account for the critical feature of tumor heterogeneity, which has been demonstrated to be closely associated with tumor aggressiveness and metastatic potential, potentially contributing to disease progression and treatment failure [13]. Tumor heterogeneity refers to the diversity of molecular biological or genetic alterations that emerge during tumor progression. However, conventional invasive detection methods (e.g., genetic analysis, molecular pathological testing, cytological evaluation, etc.) can only provide localized information about the tumor, which to some extent limits our understanding of its overall complexity. El Naqa et al. [14] first proposed the area under the curve of cumulative standardized uptake value-volume histogram (AUC-CSH) method, a quantitative approach to non-invasively characterize tumor metabolic heterogeneity (MH) based on ¹⁸F-FDG PET/CT imaging. By analyzing the spatial distribution characteristics of radiopharmaceuticals, this method can effectively reflect glucose metabolic activity and other pathophysiological processes of tumor tissues. Multiple studies [15–19] have demonstrated its significance for predicting treatment response and prognosis in various solid malignancies. Notably, it also has exhibited preliminary but encouraging prognostic potential in lymphoma [20–22]. Currently, research on the specific prognostic impact of tumor MH in DLBCL remains limited. Tumor MH may provide a novel perspective on prognosis stratification for DLBCL patients. Therefore, further validation and exploration are still necessitated to determine the optimal method for predicting the prognosis of DLBCL patients utilizing tumor MH.

This study aims to (1) assess the prognostic significance of tumor MH, as determined by ¹⁸F-FDG PET/CT imaging, on progression-free survival (PFS) and OS in patients with DLBCL receiving R-CHOP or R-CHOP-like first-line treatment regimens; (2) construct a simple, economical, and stable comprehensive prediction model; and (3) evaluate the model’s calibration, discrimination, and clinical utility and employ the bootstrap method to conduct internal validation.

Materials and methods

Study population

A total of 297 individuals with DLBCL who underwent baseline ¹⁸F-FDG PET/CT scans at the Third Affiliated Hospital of Soochow University between August 2012 and December 2022 were retrospectively analyzed in this study. The Ethics Committee of the Third Affiliated Hospital of Soochow University granted ethical approval for this study [approval number (2022) KD155], and informed consent was waived since it’s retrospective. The inclusion criteria were as follows: (1) patients with a histopathological diagnosis of DLBCL; (2) availability of a baseline ¹⁸F-FDG PET/CT scan; (3) initial treatment with R-CHOP or an R-CHOP-like regimen. Exclusion criteria included: (1) concurrent primary central nervous system lymphoma (PCNSL) or other malignancies; (2) treatment with regimens other than R-CHOP or R-CHOP-like protocols; (3) absence of baseline ¹⁸F-FDG PET/CT examination; (4) surgical removal of the primary lesion; (5) incomplete clinical data or loss to follow-up. The research strictly adheres to the principles of the Declaration of Helsinki. We collected baseline clinical data for DLBCL by reviewing electronic medical records, which included the following: age, gender, the number of extranodal involvements (EN), B symptom, Ann Arbor stage, Eastern Cooperative Oncology Group performance status (ECOG PS) score, β₂ microglobulin (β₂ MG), lactate dehydrogenase (LDH) level, cell-of-origin (COO), and NCCN-IPI score.

¹⁸F-FDG PET/CT acquisition

Image acquisition was performed using a Biograph mCT (64) PET/CT scanner (Siemens Healthcare, Erlangen, Germany). According to the standard scanning protocol, each patient fasted for at least 6 h and ensured that their blood glucose concentration was <11 mmol/L. Patients were intravenously administered ¹⁸F-FDG based on their body weight at a dose of 3.70–5.70.55MBq/kg, with a radiochemical purity of > 95%. The imaging agent was provided by Nanjing Jiangyuan Andico Positron Research and Development Co. For the process of image acquisition, each patient was required to lie on the scanning bed in a supine position with both hands clasped behind their head. Following the enhance CT scans using Care dose four-dimensional (4D) technology, a PET scan was conducted using a three-dimensional acquisition mode. The acquisition range encompassed from the base of the skull to the upper end of the femur, with a 2-minute acquisition time per bed position and approximately 6–8 bed positions, and CT data was applied for attenuation correction. The acquired images from the PET and CT scans were registered and fused on the Syngo TrueD workstation. PET data were reconstructed using an iterative reconstruction algorithm (2 iterations and 21 subsets), which incorporates point spread function correction with time-of-flight information (TrueX + TOF, ultraHD-PET). Upon completion of this process, attenuation-corrected PET images, CT images, and fused PET/CT images became available for review in axial, coronal, and sagittal planes, as well as in maximum intensity projections and three-dimensional cine mode.

¹⁸F-FDG PET/CT image analysis

Two experienced nuclear medicine physicians independently analyzed the images using Lifex software (version 5.10, http://www.lifexsoft.org), without knowledge of the patient’s clinical outcomes or other information. For disputed results, the final diagnosis was determined by a consensus among the reviewing physicians. Based on a fixed threshold of maximum standardized uptake value (SUVmax) 4.0, the system automatically generates the boundary contour of the target lesion and calculates its metabolic volume. Visual assessment was performed to manually remove physiological uptake in the brain, heart, bladder, and kidneys. Additionally, false-positive lesions resulting from inflammation or infection were also excluded. This process ensured that only true pathological lesions were retained for further analysis. Traditional semi-quantitative PET/CT parameters, including SUVmax, mean standardized uptake value (SUVmean), and total MTV (TMTV) were measured, and TLG was calculated by multiplying SUVmean by TMTV. The tumor MH of the hottest lesion in each patient was assessed using the AUC-CSH method, consistent with previous studies [20, 22, 23]. The CSH is obtained by plotting the volumetric percentage of the tumor that exceeds a certain SUVmax threshold, which ranges from 0% to 100%. AUC-CSH is a quantitative index of tracer uptake distribution heterogeneity in which lower values correspond to higher MH. All these indicators were independently measured by two physicians, and the final values were established by averaging their measurements.

Follow-up

Survival status, including disease progression, relapse, or death, and survival time were determined by reviewing electronic medical records or telephone calls, and the follow-up period concluded on August 31, 2024. Patients were followed up for a minimum of 12 months or until death. Since PFS and OS are considered reliable statistical endpoints in survival analysis, they were selected as the primary endpoints for this study. PFS was defined as the period from the baseline PET/CT scan to the first disease relapse or progression, death from any cause, or the end of follow-up. OS was defined as the time from the baseline PET/CT scan to death from any cause or the end of follow-up. Patients who did not experience an event by the end of follow-up were censored.

Development, evaluation, and validation of the nomogram

Based on the receiver operating characteristic (ROC) curve of PFS and OS, the optimal cutoff values for continuous variables including TMTV, TLG, SUVmax, SUVmean, and AUC-CSH were determined. The study also checked the multicollinearity for each variable using the variance inflation factor (VIF), and only variables with a VIF < 10 was included in the final model. Subsequently, Cox regression analysis was employed to identify independent prognostic factors associated with PFS and OS, upon which predictive models were constructed to evaluate patient survival outcomes. Variables with a P value < 0.1 in the univariate Cox regression analysis were incorporated into the multivariate Cox regression analysis. Based on the multivariate Cox proportional hazards regression model, we constructed a combined prediction model incorporating independent predictors. Calibration curves were used to graphically assess the calibration of the model, while the Harrell concordance index (C-index) and time-dependent area under the ROC curve (tdROC) were calculated to quantify its discrimination. The performance of the prediction model was subsequently compared with that of the conventional NCCN-IPI model. Decision curve analysis (DCA) was employed to assess the model’s clinical utility by calculating the net benefit at various threshold probabilities. The final model’s predictive performance was internally validated using the bootstrap resampling. The simplified analysis flowchart for the study is exhibited in Fig. 1.

Fig. 1.

Simplified analysis flowchart for the study

Statistical analysis

All statistical analyses were performed using the R software package (version 4.4.2) and IBM SPSS (version 24.0). Categorical variables were expressed as frequencies (percentages), while continuous variables were represented by mean ± standard deviation (SD) or median (minimum, maximum). The Kolmogorov-Smirnov test was employed to evaluate the normality of the variables. To analyze the consistency of measurements between two observers, a random sample of 145 cases was selected for repeated measurements, and the intraclass correlation coefficient (ICC) was used for analysis. Cox regression analysis was then applied to identify potential independent predictors and to establish a predictive model. Kaplan-Meier analysis was employed to estimate survival distributions in different risk groups, and the log-rank test was used to compare survival probability. A P value less than 0.05 was considered statistically significant.

Results

Patient characteristics

Among the 297 patients who met the inclusion criteria, there were 141 (47.5%) males and 156 (52.5%) females. The patients’ ages ranged from 13 to 87 years, with an average age of 60.63 ± 13.92 years. The median follow-up time was 32 months, during which 130 (43.8%) patients experienced disease progression and 89 (30.0%) patients died. The 1-year, 3-year, and 5-year PFS and OS rates were 73.4%, 57.9%, 54.1%, and 87.2%, 72.7% and 69.6%, respectively. 25 (8.4%) patients with missing β₂ MG values, which were replaced using the mean value imputation. The baseline clinical data and semi-quantitative PET/CT metabolic parameters of the study population are summarized in Table 1. The statistical results demonstrated good consistency for all measurements between the two observers, with ICC ranging from 0.840 to 0.999 (see Additional file for details).

Table 1.

Baseline clinical data and semi-quantitative PET/CT metabolic parameters of the study population

Character	Value
Gender, n (%)
Male	141 (47.5)
Female	156 (52.5)
Age (years), n (%)
Mean ± SD	60.63 ± 13.92
≤60	132(44.4)
>60	165 (55.6)
Ann Arbor stage, n (%)
I-II	133 (44.8)
III-IV	164 (55.2)
B symptom, n (%)
yes	79 (26.6)
no	218 (73.4)
*EN, n (%)
≤ 1	222 (74.7)
> 1	75 (25.3)
LDH, n (%)
Normal	166 (55.9)
Elevated	131 (44.1)
β2 MG, n (%)
Normal	232 (78.1)
Elevated	65 (21.9)
ECOG PS, n (%)
< 2	226 (76.1)
≥ 2	71 (23.9)
NCCN-IPI group, n (%)
Low risk (0–1)	41 (13.8)
Low−intermediate risk (2–3)	123 (41.4)
Intermediate−high risk (4–5)	105 (35.4)
High risk (≥ 6)	28 (9.4)
COO, n (%)
GCB	93 (31.3)
non-GCB	204 (68.7)
PET metabolic parameters
TMTV (cm3)
Median (range)	133 (1, 4990)
≤ 121.5, n (%)	142 (47.8)
> 121.5, n (%)	155 (52.2)
TLG (g)
Median (range)	1590 (2,32400)
≤ 1508.5, n (%)	147 (49.5)
> 1508.5, n (%)	150 (50.5)
SUVmax (g/ml)
Mean ± SD	32.6 ± 15.7
≤ 17.2, n (%)	55 (18.5)
> 17.2, n (%)	242 (81.5)
SUVmean (g/ml)
Mean ± SD	10.8 ± 4.45
≤ 4.2, n (%)	20 (6.7)
> 4.2, n (%)	277 (93.3)
AUC-CSH
Mean ± SD	0.38 ± 0.10
≤ 0.32, n (%)	93 (31.3)
> 0.32, n (%)	204 (68.7)

SD, standard deviation; EN, extranodal involvement; ^* Involvement of major organs (bone marrow, central nervous system, liver, gastrointestinal tract, or lung); LDH, lactate dehydrogenase; β₂ MG, β₂microglobulin; ECOG PS, eastern cooperative oncology group performance status; NCCN-IPI, National Comprehensive Cancer Network-international prognostic index; COO, cell-of-origin; GCB, germinal center B-cell; SUVmax, maximum standardized uptake value; SUVmean, mean standardized uptake value; TMTV, total metabolic tumor volume; TLG, total lesion glycolysis; AUC-CSH, area under the curve of cumulative standardized uptake value-volume histogram. The cutoff value of semi-quantitative PET metabolic parameters is the optimal threshold for predicting the PFS endpoint event

After conducting a multicollinearity screening for all variables, it was found that the VIF value of SUVmax >10, indicating the existence of multicollinearity. We then exclude SUVmax from subsequent analyses to maintain the stability and reliability of our regression model. ROC curves were employed to determine the optimal cutoff values for predicting PFS and OS of each marker. The values of SUVmean, TMTV, TLG, and AUC-CSH are 4.2 g/ml, 121.5 cm³, 1508.5 g, and 0.32 for PFS, whereas they are 4.15 g/ml,156.5 cm³and 1527.5 g, and 0.32 for OS, as detailed in Table 2.

Table 2.

Determining the optimal cutoff values for semi-quantitative PET/CT metabolic parameters based on ROC curves

	PFS		OS
Character	AUC	cutoff	AUC	cutoff
TMTV (cm3)	0.707	121.5	0.728	156.5
TLG (g)	0.679	1508.5	0.702	1527.5
SUVmean (g/ml)	0.482	4.2	0.451	4.15
AUC-CSH	0.639	0.32	0.639	0.32

PFS progression-free survival, OS overall survival, TMTV total metabolic tumor volume, TLG total lesion glycolysis, SUVmean mean standardized uptake value, AUC-CSH area under the curve of cumulative SUV-volume histogram

Patients with a high degree of tumor MH experience poorer survival outcomes

The AUC-CSH values of this study ranged from 0.14 to 0.72, with a mean ± SD of 0.38 ± 0.10. Generally, a lower value of AUC-CSH indicates a higher level of tumor heterogeneity and poorer survival outcomes. Patients with a high degree of tumor MH (AUC-CSH ≤ 0.32) exhibited significantly lower 1-year, 3-year, and 5-year PFS rates (65.6% vs. 77.0%, 41.4% vs. 65.4%, and 34.3% vs. 62.9%, respectively; P < 0.0001) and OS rates (82.8% vs. 89.2%, 62.8% vs. 77.3%, and 56.4% vs. 75.6%, respectively; P = 0.00032) compared to those with low tumor MH (AUC-CSH > 0.32), as shown in Additional file. Furthermore, Pearson correlation analysis revealed a weak or very weak negative correlation between AUC-CSH and the MTV of the hottest lesion (r = −0.189, P = 0.001) or TMTV (r = −0.263, P < 0.0001). Suggesting that tumor MH is independent of MTV, which represents tumor burden. These results are detailed in the Additional file.

Univariate and multivariate cox regression analysis

Univariate Cox regression analysis indicated that age, Ann Arbor stage, B symptom, EN, ECOG PS, LDH, TMTV, TLG, SUVmean, and AUC-CSH are potential prognostic factors for PFS (all P < 0.05). However, no significant correlation was observed between PFS and gender, β₂ MG, or COO (all P > 0.05). Similarly, OS significantly correlated with age, Ann Arbor stage, B symptom, EN, ECOG PS, LDH, β₂MG, COO, TMTV, TLG, and AUC-CSH (all P < 0.05), while gender and SUVmean showed no significant correlation with OS (all P > 0.05), as presented in Table 3. Statistically significant variables were incorporated into a multivariate Cox regression analysis, revealing that only age (hazard ratio [HR], 1.62, 95% confidence interval [CI], 1.13 ~ 2.34, P = 0.01), TMTV (HR, 3.24, 95% CI, 1.56 ~ 6.73, P = 0.002), and AUC-CSH (HR, 0.64, 95% CI, 0.45 ~ 0.92, P = 0.015) remained as independent prognostic factors for PFS. Additionally, age (HR, 2.15, 95% CI, 1.35 ~ 3.42, P = 0.001), TMTV (HR, 3.35, 95% CI, 1.30 ~ 8.62, P = 0.012), ECOG PS (HR, 1.65, 95% CI, 1.03 ~ 2.64, P = 0.04), COO (HR, 0.50, 95% CI, 0.28 ~ 0.86, P = 0.013), and AUC-CSH (HR, 0.59, 95% CI, 0.38 ~ 0.92, P = 0.018) were identified as independent prognostic factors for OS.

Table 3.

Univariate and multivariate Cox regression analysis of progression-free survival and overall survival between baseline clinical characteristics and semi-quantitative PET/CT metabolic parameters

Character	PFS				OS
	Univariate analysis		Multivariate analysis		Univariate analysis		Multivariate analysis
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
Age, ≤ 60/>60	1.55 (1.09, 2.22)	0.015*	1.62 (1.13, 2.34)	0.01*	1.9 (1.27, 3.11)	0.003*	2.15 (1.35, 3.42)	0.001*
Gender, Male/Female	1.17 (0.83, 1.65)	0.382	—	—	1.04 (0.69,1.58)	0.846	—	—
Ann Arbor stage, I-II/III-IV	1.57 (1.10, 2.25)	0.013*	1.11 (0.74, 1.68)	0.61	1.61 (1.04,2.50)	0.031*	1.04 (0.64, 1.70)	0.86
B symptom, no/yes	1.50 (1.03, 2.18)	0.033*	1.05 (0.70, 1.56)	0.82	2.09 (1.36, 3.21)	< 0.001*	1.20 (0.75, 1.94)	0.45
EN, ≤1/>1	1.82 (1.27, 2.62)	0.001*	1.14 (0.76, 1.70)	0.53	2.70 (1.78, 4.12)	< 0.001*	1.57 (0.97, 2.55)	0.07
ECOG PS,0–1/≥2	1.97 (1.36, 2.84)	< 0.001*	1.26 (0.85, 1.87)	0.25	3.10 (2.03, 4.72)	< 0.001*	1.65 (1.03,2.64)	0.04*
LDH, normal/elevated	1.76 (1.24, 2.48)	0.001*	1.39 (0.94, 2.06)	0.10	1.94 (1.27, 2.96)	0.002*	1.50 (0.93,2.42)	0.10
β2 MG, normal/elevated	1.37 (0.92, 2.02)	0.117	—	—	1.74 (1.11, 2.74)	0.016*	0.97 (0.59, 1.59)	0.89
COO, non-GCB/GCB	0.72 (0.49, 1.07)	0.107	—	—	0.47 (0.28, 0.80)	0.005*	0.50 (0.28, 0.86)	0.013*
TMTV, low/high	3.41 (2.32, 5.02)	< 0.001*	3.24 (1.56, 6.73)	0.002*	3.99 (2.50, 6.39)	< 0.001*	3.35 (1.30, 8.62)	0.012*
TLG, low/high	2.77 (1.91, 4.01)	< 0.001*	0.74 (0.36, 1.52)	0.41	3.20 (2.01, 5.09)	< 0.001*	0.67 (0.26, 1.71)	0.40
SUVmean, low/high	2.85 (1.05, 7.72)	0.039*	1.44 (0.51, 4.07)	0.49	2.02 (0.64, 6.39)	0.231	—	—
AUC-CSH, low/high	0.50 (0.35, 0.71)	< 0.001*	0.64 (0.45, 0.92)	0.015*	0.47 (0.31, 0.72)	< 0.001*	0.59 (0.38, 0.92)	0.018*

PFS progression-free survival, OS overall survival, EN extranodal involvement, LDH lactate dehydrogenase, β₂ MG β₂ microglobulin, ECOG PS Eastern Cooperative Oncology Group performance status, COO cell-of-origin, GCB germinal center B-cell, SUVmean mean standardized uptake value, TMTV total metabolic tumor volume, TLG total lesion glycolysis, AUC-CSH area under the curve of cumulative SUV-volume histogram, CI confidence interval; ^* is refers to the P value < 0.05

Development and validation of the combined prediction model

Variables with independent prognostic significance in the multivariate Cox regression analysis were utilized to construct a combined prediction model. Given that ECOG PS and COO did not exhibit in dependent prognostic value for PFS, we chose age, TMTV, and AUC-CSH as indicators to construct a combined model for individualized prediction of PFS and OS at 1-year, 3-year, and 5-year (visualized as a nomogram, as shown in Fig. 2a and d). Each predictive variable was assigned a score based on a predefined scoring system. By summing the scores of these variables and mapping the total score to a survival probability scale, we were able to estimate the probabilities of 1-year, 3-year, and 5-year PFS and OS. The accuracy of the combined model in predicting 1-year, 3-year, and 5-year PFS and OS was evaluated by ROC curves, with AUC values of 0.748, 0.744, and 0.741 for PFS and 0.746, 0.775, and 0.751 for OS, respectively (Fig. 3). Figure 2b and e exhibit the calibration curves for predicting the probability of PFS and OS at 1-year, 3-year, and 5-year for the combined model after 1000 bootstrap repetitions, which showed good consistency between the prediction and actual observations. As shown in Table 4 and Fig. 2c and f, the performance of the combined model for predicting PFS and OS is superior to the NCCN-IPI model, with C-index of 0.688 (95% CI, 0.645 ~ 0.730), 0.630 (95% CI, 0.588 ~ 0.697) for PFS and 0.721 (95% CI, 0.670 ~ 0.772), 0.688 (95% CI, 0.641 ~ 0.734) for OS, respectively.

Fig. 2.

Nomogram and calibration curves for predicting 1-year, 3-year, and 5-year PFS (a, b) and OS (d, e) integrating age, TMTV, and AUC-CSH. The gray solid line signifies the ideal scenario where the predicted survival rates perfectly align with the actual survival rates; the blue, orange, and purple solid lines represent the actual predicted survival rates for PFS and OS at 1-year, 3-year, and 5-year, respectively. The closer these lines are to the diagonal gray line, the better the predictive ability. Figure 2c and f present the tdROC for predicting PFS and OS using the combined (blue line) and NCCN-IPI models (orange line), respectively

Fig. 3.

ROC curves and the areas under the curve of the combined model for predicting 1-year, 3-year, and 5-year PFS (a) and OS (b) rates

Table 4.

The results of harrell’s C-index

		C-index	95% CI
PFS	Combined model	0.688	0.645 ~ 0.730
	NCCN-IPI	0.630	0.588 ~ 0.697
OS	Combined model	0.721	0.670 ~ 0.772
	NCCN-IPI	0.688	0.641 ~ 0.734

CI confidence interval, PFS progression-free survival, OS overall survival, NCCN-IPI National Comprehensive Cancer Network-international prognostic index

To evaluate the accuracy and robustness of the combined prediction model, we conducted internal validation using the bootstrap method, performing 1000 resamples and generating ROC curves. The results demonstrated an accuracy of 0.6552376, a concordance of 0.2957067, and an AUC of 0.714 for PFS, and 0.7204392, 0.253757, 0.749 for OS. As illustrated in Fig. 4, the light blue curves are clustered closely around the dark blue curve, suggesting that the model’s performance is relatively stable.

Fig. 4.

The ROC curves for internal validation for predicting PFS and OS using the bootstrap method. The light blue curves represent the ROC curves obtained from different sampled datasets using the Bootstrap method, while the dark blue solid line represents the ROC curves of the combined prediction model for PFS (a) and OS (b)

Performance analysis and clinical practicability of the combined model

The DCA of the combined model is depicted in Fig. 5. The curve demonstrates that the combined model provides a superior net benefit in predicting both PFS and OS compared to the “no-treatment” or “treatment-all” strategies. This indicates its potential practicality in clinical decision-making. For PFS, the combined model yielded a higher overall net benefit compared to the NCCN-IPI model across all the range of risk thresholds. While for OS, the combined model also provided a greater overall net benefit than the NCCN-IPI model across the majority of risk thresholds.

Fig. 5.

DCA of the combined and NCCN-IPI model for predicting1-year, 3-year, and 5-year PFS (a) and OS (b), in which the blue solid line represents the combined model and the purple solid line represents the NCCN-IPI model

Kaplan-Meier survival curves for PFS and OS in DLBCL patients, stratified by the total scores generated from nomogram

According to the optimal cutoff value for the total scores from nomograms in terms of PFS and OS, all patients were categorized into three different risk groups, whose distribution is as follows: low-risk group (total score < 39.8 for PFS, < 53.7 for OS), intermediate-risk group (total score 39.8 ~ 139.8 for PFS, 53.7 ~ 153.7 for OS), and high-risk group (total score > 139.8 for PFS, > 153.7 for OS).

The median PFS and OS for the high-risk group were 13 and 28 months, respectively. As shown in Table 5, the 1-year, 3-year, and 5-year PFS rates in the high-risk group were 51.7%, 33.7%, and 27.8%, respectively, while the 1-year, 3-year, and5-year OS rates were 71.8%, 46.0%, and 44.1%, respectively. The Kaplan-Meier survival curves showed that statistically significant differences in PFS (Fig. 6a) and OS (Fig. 6b) were observed among the three risk groups (P < 0.0001), and patients in the high-risk group had significantly worse survival outcomes compared with those in the low-risk group both for PFS and OS (P < 0.0001). The typical cases were shown in Fig. 7.

Table 5.

Median survival time and 1-year, 3-year, and 5-year PFS and OS rates for the three risk groups stratified by the total scores generated from the nomogram

	Risk group (total score)	N	Median survival time (month)	1-year	3-year	5-year
PFS	Low-risk: >39.8	64	n.r	93.8%	80.7%	76.3%
	Intermediate-risk: 39.8–139.8	146	n.r	77.4%	62.0%	60.0%
	High-risk: >139.8	87	13	51.7%	33.7%	27.8%
OS	Low-risk: <53.7	70	n.r	97.1%	92.7%	89.2%
	Intermediate-risk: 53.7–153.7	149	n.r	90.6%	77.2%	73.3%
	High-risk: >153.7	78	28	71.8%	46.0%	44.1%

N the number of patients, n.r. not reached, PFS progression-free survival, OS overall survival

Fig. 6.

The Kaplan-Meier survival curves of the three risk groups stratified by the total scores generated from the nomogram for predicting PFS (a) and OS (b)

Fig. 7.

Typical cases of predicting PFS and OS using a combined model incorporating age, TMTV, and AUC-CSH. Case A: A 65-year-old female patient (score 40 for PFS, 53.75 for OS) presenting with a TMTV of 105.5cm³ (score 0 for both PFS and OS) and an AUC-CSH of 0.5 (score 0 for both PFS and OS). According to the nomogram, the total scores are 40 for PFS and 53.75 for OS, predicting a 5-year PFS rate of > 70% and a 5-year OS rate of > 80%. The patient achieved complete remission after completing first-line therapy and maintained disease remission for 82 months, with no events of relapse, progression, or death observed throughout the follow-up period. Case B: A 74-year-old male patient (score 40 for PFS, 53.75 for OS) with a TMTV of 341cm³(score 100 for both PFS and OS) and an AUC-CSH of 0.26 (score 38.75 for PFS, 35 for OS). The hottest lesion is indicated by the red arrow. The total scores are 178.75 for PFS and 188.75 for OS based on the nomogram, predicting a 5-year PFS rate of 20% and a 5-year OS rate of < 35%. The patient achieved partial remission from the end of standard first-line treatment and disease progression one year later, leading to the patient’s death due to the illness at 21 months

Discussion

This study confirmed the predictive significance of tumor MH, assessed by ¹⁸F-FDG PET/CT, for patients newly diagnosed with DLBCL. Survival analysis identified age, TMTV, and tumor MH as independent predictors of both PFS and OS. A combined model constructed based on these three factors stratified patients into three risk groups, among which the high-risk group exhibited 5-year PFS and OS rates of only 27.8% and 44.1%, respectively.

Our study systematically analyzed the independent predictive value of tumor MH in the prognosis of DLBCL and evaluated its clinical application potential. Patients exhibiting higher degree of tumor MH had a significantly worse prognosis (PFS: HR, 0.64, 95% CI, 0.45 ~ 0.92, P = 0.015 and OS: HR, 0.59, 95% CI, 0.38 ~ 0.92, P = 0.018). A prospective study demonstrated that both tumor MH (HR, 12.8; 95% CI: 3.3 ~ 49.9, P < 0.001) and TLG (HR, 46.5; 95% CI: 5.8 ~ 373.8, P < 0.001) were independent predictors of PFS. Based on baseline tumor MH and TLG, the investigators constructed a prognostic model for PFS, which confirmed that patients with both high TLG and high MH had significantly worse clinical outcomes (HR, 31.2; 95% CI, 9.9 ~ 97.8, P < 0.001), with a 5-year PFS rate of only 11% (95% CI, 1 ~ 39%). This study is the first to establish the clinical significance of tumor MH as a potential prognostic biomarker in primary mediastinal B-cell lymphoma (PMBCL) [20]. Subsequent studies conducted by Senjo H et al. and Ceriani Let al. similarly confirmed the adverse prognostic impact of higher degree of tumor MH in DLBCL, further establishing the status of tumor MH as a prognostic biomarker for DLBCL [21, 22]. Although tumor MH did not persist as a significant risk factor in multivariate analysis, it remained valuable for stratifying patients with high TMTV. This observation may be attributed to the more prominent prognostic significance of TMTV compared with MH. Our data and the findings of the previous studies collectively support the viewpoint that tumor MH holds promise as a biomarker for individualized prognostic prediction and treatment stratification in DLBCL patients.

Tumor heterogeneity may be one of the underlying mechanisms contributing to increased tumor aggressiveness and therapeutic resistance, thereby partly explaining inter-individual differences in treatment response [24, 25]. Reinfeld BI et al. [26] founded that ¹⁸F-FDG is taken up not only by tumor cells but also by immune infiltrating cells within tumor lesions. Consequently, ¹⁸F-FDG uptake heterogeneity can more comprehensively reflect the heterogeneity of both tumor cells and immune cells within tumor lesions. Analysis of intra-tumoral metabolic activities enables more precise characterization of tumors, providing profound imaging evidence for clinical treatment and aiding in our understanding of this complex phenomenon. Nevertheless, these results necessitate verification via pertinent mechanistic studies.

TMTV serves as a robust predictor of prognosis in DLBCL and other lymphoma subtypes, regardless of the measurement methodology employed., which align with our finding [10, 11, 27]. PET-derived TMTV offers a more thorough assessment of disease burden than anatomical imaging [28–31], as it incorporates both metabolic activity and anatomical features of the tumor. Compared to traditional prognostic models - which indirectly reflect tumor burden through Ann Arbor stage and LDH levels-MTV provides a more accurate representation of tumor burden and may even partially replace these parameters [31, 32]. This could potentially explain why, in our study, apart from age, other clinical indicators within the NCCN-IPI failed to demonstrate significant prognostic value in multivariate analysis. Conversely, Mikhaeel NG et al. [10]. found that the Ann Arbor stage retained independent prognostic value, providing additional prognostic information beyond MTV alone. This discrepancy may stem from the variations in the methodology. They added age and stage as continuous variables to the international metabolic prognostic index (IMPI) model, unlike the dichotomous variables applied in our study. A further reason for the inconsistent results could be the difference in the proportion of advanced disease (Ann Arbor stage III–IV): 65.2% in their cohort versus 55.2% in ours. Our study exhibited a more balanced stage distribution, while their cohort consisted predominantly of advanced stage patients, contributing to an overall unfavorable outcome and a relatively higher number of events. The Hans algorithm-based classification into GCB and non-GCB subtypes revealed no significant association with PFS. This finding underscores the limitations of the Hans classification as a standalone tool for risk stratification in DLBCL. Unlike MTV, tumor MH highlights information such as spatial distribution, heterogeneity, and morphological characteristics. We found that tumor MH is independent of both the MTV of the hottest lesion and the TMTV. These two metrics are complementary, and their combination can provide a more comprehensive prognostic evaluation in DLBCL.

It is common knowledge that age serves as a crucial indicator reflecting the biological reserve function of cancer patients. Elderly individuals typically exhibit reduced tolerance to treatments such as chemotherapy and radiotherapy, and they are more likely to experience adverse effects from these therapies. MTV was reported to be highly correlated with all factors in IPI except for age [28]. By incorporating age into the combined model, the prediction accuracy of the model can be further improved.

In the current study, we had several limitations. First, its retrospective nature may introduce potential selection bias. Prospective, multicenter and large-sample studies are still necessary to support the prognostic significance of tumor MH in DLBCL. Second, the analysis relied on a single dataset with a relatively small sample size, which may affect the generalizability of the findings. The absence of external dataset verification also restricts the model’s applicability in real-world settings. Third, maximum distance between lesions (Dmax), an emerging indicator which reflects the spatial distribution of the disease, provided prognostic information independent of MTV [33, 34]. However, it has not yet been routinely adopted in clinical practice. We will quantitatively measure the Dmax in the future to evaluate its application in DLBCL. Moreover, incorporating additional prognostic biomarkers such as the incidence of double-hit or triple-hit lymphomas and the expression frequencies of MYC, BCL2, and BCL6, as well as interim PET (iPET) data (e.g., Deauville score (DS) or ΔSUVmax) remains valuable for DLBCL prognostic. Although these features were not analyzed in the current study due to data limitations. In a follow-up study, multiple prognostic indicators will be integrated into the model to identify those with the greatest impact on clinical outcomes to assist in further risk stratification of DLBCL and guide individualized treatment decisions. Last, further integration of radiomics and advanced artificial intelligence (AI) - based image analysis may also help build a more robust and comprehensive prognostic evaluation system.

Conclusions

Tumor MH holds promise as a biomarker for individualized prognostic prediction and treatment stratification in DLBCL patients. By revealing the information regarding the intralesional heterogeneity and spatial complexity, tumor MH complements DLBCL tumor burden measured by TMTV.

Abbreviations

MH

Metabolic heterogeneity

¹⁸F-FDG PET/CT

18-Fluorine fluorodeoxyglucose positron emission tomography computed tomography

DLBCL

Diffuse large B-cell lymphoma

SUVmax

Maximum standardized uptake value

SUVmean

Mean standardized uptake value

TMTV

Total metabolic tumor volume

TLG

Total lesion glycolysis

AUC-CSH

Area under the curve of cumulative standardized uptake value-volume histogram

HR

Hazard ratio

CI

Confidence interval

PFS

Progression-free survival

OS

Overall survival

NCCN-IPI

National Comprehensive Cancer Network-International prognostic index

DCA

Decision curve analysis

R-CHOP

Rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone

R/R

Relapsed or refractory

EN

Extranodal involvement

β₂ MG

β₂ microglobulin

LDH

Lactate dehydrogenase

ECOG PS

Eastern Cooperative Oncology Group performance status

COO

Cell-of-origin

ROC

Receiver operating characteristic

VIF

Variance inflation factor

tdROC

Time-dependent area under the receiver operating characteristic curve

C-index

Concordance index

PCNSL

Primary central nervous system lymphoma

SD

Standard deviation

ICC

Intraclass correlation coefficient

GCB

Germinal center B-cell

PMBCL

Primary mediastinal B-cell lymphoma

IMPI

International metabolic prognostic index

Dmax

Maximum distance between two lesions

iPET

Interim positron emission tomography

Author contributions

This study was conceived and designed by WCX and FW under the supervision and guidance of YTW and WYG. WCX collected and analyzed information from the PET/CT image data. FW was responsible for the collection of clinical information and prognosis follow-up. LL, HYZ, and BH assisted in collecting follow-up data. The results were analyzed by WCX, who also edited the first manuscript. YTW and WYG reviewed and revised the manuscript. All authors have read and approved the final version.

Funding

This work was supported by Clinical Medical Science and Technology High-end Platform and Transformation Base Construction Project of Soochow University (Characteristic Discipline)-Nuclear Medicine (PI: Yuetao Wang), Changzhou Clinical Medical Center (Nuclear Medicine) (CZZX202204, PI: Yuetao Wang), Outstanding Talent of Changzhou “The 14th Five-Year Plan” High-Level Health Talents Training Project (2022 − 260, PI: Yuetao Wang).

Data availability

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

Declarations

Ethics approval and consent to participate

The study protocol adhered to the principles of the Declaration of Helsinki and received approval from the ethics committee of the Third Affiliated Hospital of Soochow University. Given the nature of this study being retrospective, the requirement for written informed consent was exempted.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.