The utility of 18F-FDG PET/CT in assessing bone marrow involvement and prognosis in newly diagnosed diffuse large B-cell lymphoma

A retrospective analysis was conducted on DLBCL patients who underwent PET/CT scans between 2017 and 2024. These patients had not undergone radiotherapy, chemotherapy, hematopoietic factor therapy, or had a previous history of other malignancies prior to the PET/CT scans. Patient confidentiality was maintained in compliance with national regulations. The study was approved by the Ethics Committee of the Second Affiliated Hospital of Harbin Medical University (Approval No. YJSKY2023-327, 2023-09-04).

Clinical data was sourced from institutional information systems. A review was conducted on the following clinical parameters: age, sex, Eastern Cooperative Oncology Group performance status (ECOG PS), B-symptoms, clinical manifestations (pain, nosebleeds, loss of appetite, bloating, among others), and extranodal involvement. The baseline laboratory findings obtained within a 2-week window before and after PET/CT imaging encompassed serum lactate dehydrogenase (LDH), lymphocyte-to-monocyte ratio (LMR > 3), neutrophil-to-lymphocyte ratio (NLR > 2.5)⁷, leukopenia (< 4×10⁹/L), anemia (hemoglobin < 120g/L), and thrombocytopenia (platelet counts < 100×10⁹/L). The International Prognostic Index (IPI) score was based on these data. Following PET/CT scans, all patients underwent first-line therapy. The study cohort was derived according to the process detailed in Supplementary Figure 1.

Imaging was performed using a PET/CT system (Biograph64 mCT, Siemens Healthcare, Berlin, Germany) with a full-ring PET scanner. Patients fasted for 6 hours and maintained blood glucose levels below 11 mmol/L before the procedure. After FDG injection (0.11 mCi/kg), patients rested supine for 60 minutes. CT was performed with a 120 kV tube voltage and modulated tube current (CARE Dose). PET acquisition was conducted at 1.6 mm/s, and images were reconstructed with the TrueD algorithm for attenuation correction using CT data. The PET matrix was 200×200 (voxel size: 4.07×4.07×3 mm), and the CT matrix was 512×512 (voxel size: 0.78×0.78×1 mm), with PET images reconstructed using the TrueX + TOF method.

Image interpretation was conducted through visual and semi-quantitative analysis, utilizing glucose activity in the normal liver as a reference to assess the presence of BMI. In cases where BM lesions were identified, they were classified as focal lesions (demonstrating elevated FDG uptake in one or more localized regions) or diffuse lesions (exhibiting a uniform increase in FDG uptake across the entire marrow space). These lesions, with or without associated bone destruction, were distinguished from benign findings based on standard CT images or medical history (e.g., fractures). In such cases, BM positivity on PET/CT was ascertained. The maximum standardized uptake value (SUVmax) of the liver was measured in a 2-cm region of interest in the right lobe, while the highest SUVmax for BM was assessed at lumbar vertebrae 1-5 in cases of diffuse uptake. The BM-to-liver SUVmax ratio (BLR) was calculated to investigate survival outcomes. Two experienced nuclear medicine specialists, each possessing more than 5 years of professional experience, independently interpreted the images, with discrepancies resolved by consensus (Figure 1).

FIGURE 1.

The initial PET maximum-intensity projection images, and lymph nodes pathology confirmed DLBCL with normal (A), focal increased in the right humerus (B), diffuse increased (C, D).

All patients underwent unilateral iliac BM core needle biopsy and aspiration without imaging guidance. BMB samples were collected within one month before or after the PET/CT imaging and evaluated by a senior hematologist.

If BM uptake paralleled the uptake activity of other lymphoma lesions during follow-up with PET/CT or was confirmed by BMB, the diagnosis of BMI was confirmed.^8,9

Receiver operating characteristic (ROC) curve analysis was employed to establish the optimal BLR threshold for survival analysis. Progression-free survival (PFS) was defined as the duration from the definitive diagnosis to the first occurrence of disease progression or relapse, all-cause mortality, or the most recent follow-up. The Shapiro-Wilk test was utilized for assessing the distribution of continuous variables in the baseline data. Continuous variables were expressed as median and interquartile range. Categorical data were presented as patient counts and percentages. Disparities in variables between the two groups were assessed using logistic regression analysis and the chi-square test. Comparison of SUVmax and BLR was conducted using the Kruskal-Wallis test and Dunn’s post-hoc test. Cox proportional hazards regression models were utilized to analyze the impact of factors such as IPI and related components, BLR, and BM status on survival outcomes. Survival analysis was performed using Kaplan-Meier curves and log-rank tests. Statistically significant variables (p < 0.05) from the univariate analysis were entered into the multivariate Cox regression. The resulting independent prognostic factors were incorporated to construct a prognostic nomogram. The model was validated internally using bootstrap resampling with 1,000 iterations. The discriminative ability of the nomogram was quantified by the concordance index (C-index), and its accuracy was visually assessed with a calibration curve comparing predicted probabilities against observed frequencies. Statistical analyses were performed using IBM SPSS 27 and R 4.3.3 software.

In this study, a total of 57 DLBCL cases were analyzed in Table 1. The cohort showed 27 males (47.4%) and 30 females (52.6%), with a median age of 61 years old (range: 28–81). Among 57 patients, 30 (52.6%) were aged 60 years or older.

The baseline characteristics of the patients defined by the two diagnostic methods – (1) the PET/CT-based cohort and (2) the BMB-based cohort – were compared. There were no significant differences between the two groups in terms of sex, age, LMR > 3, NLR > 2.5, clinical symptoms, B-symptoms, LDH (+), ECOG PS ≥ 2, extranodal sites ≥ 2, or IPI score > 2. However, there were statistical differences in the anemia, BM SUVmax, and BLR levels (all p < 0.05).

Out of the 34 patients detected as BMI positive by PET/CT, 32 were confirmed as true positive cases (20 confirmed via BMB, 12 through follow-up). Among the 23 patients identified as BMI negative by PET/CT, 2 were verified as false negative by BMB, while the remaining 21 were confirmed as negative by BMB or follow-up assessments (Table 2).

TABLE 2.

We identified 34 patients (59.6%) with BMI and 23 (40.4%) without BMI. Using the final clinical diagnosis as the reference standard, PET/CT correctly identified 32 (56.1%) true positive cases, while BMB correctly identified 20 (35.1%) true positive cases. The diagnostic results from PET/CT and BMB were consistent in 39 cases. Specifically, 18 (31.6%) demonstrated concordant positive findings on both BMB and PET/CT tests, while 21 (36.8%) displayed negative outcomes for BMB and PET/CT. However, there were 18 (31.6%) cases where the diagnostic results between PET/CT and BMB were inconsistent, including 16 (28.1%) that tested positive for PET/CT yet negative for BMB, and 2 (3.5%) that tested negative for PET/CT but positive for BMB (Table 3).

Table 4 compares the diagnostic performance of PET/CT and BMB against the pre-defined BMI. PET/CT demonstrated superior diagnostic efficacy, with an accuracy of 93.0% and a sensitivity of 94.1%. Its specificity, positive predictive value (PPV), and negative predictive value (NPV) were 91.3%, 94.1%, and 91.3%, respectively. The high Youden index (0.854) and excellent agreement with the reference standard (kappa = 0.854) further confirm its robustness.

Utilizing the optimal cut-off value of BLR = 1.340 (AUC = 0.755, p = 0.004), patients were stratified into high- and low-BLR groups. All 27 patients (100%) in the high-BLR group were interpreted as positive by visual PET/CT assessment. In contrast, only 7 of the 30 patients (23.3%) in the low-BLR group were PET/CT-positive. The agreement between the quantitative (BLR-based) and visual PET/CT assessments was excellent (p < 0.001, kappa = 0.757).

Among all patients, the median SUVmax of 23 patients with normal BM uptake was 2.48 (2.12, 3.02), and the BLR was 0.80 (0.75, 0.87). In contrast, the median SUVmax of 18 patients with focal BM uptake was 9.36 (3.73, 13.30; p < 0.001), BLR was 3.49 (1.60, 4.67; p < 0.001), and the median SUVmax of 16 patients with diffuse BM uptake was 4.04 (3.04, 5.86; p = 0.006), BLR was 1.48 (1.23, 1.81; p < 0.001). The difference in SUVmax and BLR between focal and diffuse uptake groups was not statistically significant (Figure 2A and C). The median SUVmax and BLR of 34 patients with increased BM uptake were 1.77 (1.37, 3.68) and 4.74 (3.42, 9.43), respectively, with statistical differences (both p < 0.001). Using BMB as the standard for analysis, the median SUVmax and BLR of 37 cases of BMB (-) were 2.78 (2.28, 3.55) and 0.79 (0.89, 1.44), while 20 cases of BMB (+) had higher median SUVmax and BLR, which were 6.39 (4.47, 11.73; p < 0.001) and 1.89 (1.40, 3.78; p < 0.001), respectively. The median SUVmax of focal and diffuse BM uptake were 11.25 (8.25, 17.03) and 4.80 (3.35, 6.24), respectively, and the median BLR were 3.72 (2.15, 5.70) and 1.42 (1.09, 1.86), respectively (Figure 2B and D). The 18 patients who were positive on both PET/CT and BMB exhibited significantly elevated median SUVmax and BLR compared to the rest of the cohort: SUVmax 8.00 (4.20-12.45) vs. 2.92 (2.32-3.78), and BLR 1.92 (1.57-4.00) vs. 0.89 (0.78-1.39); all comparisons were statistically significant (p < 0.001).

FIGURE 2.

Distribution of uptake according to bone marrow biopsy (BMB) and PET/CT.

**, p = 0.006;

****, p < 0.0001; BLR = the ratio of the maximum standardized uptake values of bone marrow-to-liver; ns = statistically nonsignificant; SUVmax = maximum standardized uptake value of bone marrow

According to the Ann Arbor stage, PET/CT and BMB were independently assessed. PET/CT classified 46 patients (80.7%) as stage IV, while BMB classified 44 patients (77.2%) as stage IV. The final clinical diagnosis, which integrated all available data, led to the downstaging of two patients: one from stage to I stage IV, and another from stage III to stage IV. This demonstrates that PET/CT plays a crucial role in preventing understaging, thereby ensuring more accurate disease staging.

All 57 patients were included in the survival analysis. The median survival time was not reached at the time of analysis, with 15 patients experiencing disease progression or death. 34 patients were BMI (+), with a median PFS of 15 months and the 2-year PFS rate 47.9% ± 10.7%. In visual analysis, both a positive BMB (p = 0.016, Figure 3A) and a positive PET/CT finding (p = 0.012, Figure 3B) were significantly associated with shorter PFS; In quantitative analysis, high BLR levels (cut-off value, 1.340) were correlated with shorter PFS (p < 0.001, Figure 3C). The 2-year PFS rates were 92.6% ± 5.0% for the low-BLR group and 35.8% ± 12.7% for the high-BLR group.

FIGURE 3.

Kaplan-Meier curves for progression-free survival (PFS) in total study patients according to Eastern Cooperative Oncology Group performance status (ECOG PS) (A), bone marrow biopsy (BMB) (B), PET/CT (C) and the ratio of the maximum standardized uptake values of bone marrow-to-liver (BLR) groups (D)

Variables found to be significant in the univariate Cox analysis for PFS are listed in Table 5. Multivariate analysis confirmed that high BLR levels and an ECOG PS ≥ 2 were independent risk factors for shorter PFS (Table 5). Kaplan-Meier survival plots visually corroborated the poor prognosis associated with a high BLR and an ECOG PS ≥ 2 (Figure 3A and D).

Based on the results of the multivariate analysis, we constructed a nomogram for predicting PFS (Supplementary Figure 2). The nomogram demonstrated good predictive accuracy for PFS, with a C-index of 0.812. The calibration curve for PFS survival probabilities showed strong agreement between the actual observed probabilities and those predicted by the nomogram (Supplementary Figure 3).