Prediction of lymph node metastasis in non-small cell lung cancer and its correlation with Ki-67 expression: a comparative study between intravoxel incoherent motion imaging and ¹⁸F-FDG PET

Abstract

Background

Integrated positron emission tomography/magnetic resonance (PET/MRI) may have the potential to evaluate lymph node metastasis (LNM) status and Ki-67 proliferation index in patients with non-small cell lung cancer.

Methods

This study enrolled 92 pathologically confirmed NSCLC patients. Quantitative analysis of metabolic parameters, such as maximum standardized uptake value (SUVmax), metabolic tumor volume (MTV), total lesion glycolysis (TLG), and IVIM parameters including apparent diffusion coefficient (ADC), true diffusion coefficient (D), pseudo diffusion coefficient (D*), perfusion fraction (f), and distributed diffusion coefficient (DDC). The predictive performance of each parameter for LNM was assessed using receiver operating characteristic (ROC) curve analysis, and a multivariate logistic regression model was constructed to establish the optimal combined predictive model. Spearman correlation analysis was used to explore the relationship between imaging parameters and Ki-67 expression.

Results

LNM prediction: The LNM-positive group exhibited lower ADC, D, and DDC (all P < 0.05) compared to the LNM-negative group. Multivariate analysis identified MTV and DDC as independent predictors of LNM. The combined model (MTV + DDC) achieved an AUC of 0.821 (sensitivity 79.49%, specificity 73.59%), significantly outperforming individual parameters (P < 0.05). Ki-67 correlation: SUVmax, MTV, and TLG showed positive correlations with Ki-67 (r = 0.232 to -0.300, P < 0.05), while ADC and D values exhibited negative correlations (r = − 0.327 to − 0.240, P < 0.05). No significant association was observed between D*, f, and Ki-67 (P > 0.05).

Conclusion

Integrated PET/MRI, by combining metabolic (MTV) and diffusion (DDC) parameters, significantly improves the predictive accuracy for LNM in NSCLC. Additionally, metabolic and select IVIM parameters correlate with Ki-67 expression.

Introduction

Non-small cell lung cancer (NSCLC) accounts for approximately 85% [1] of all lung cancer cases, and its clinical prognosis is closely associated with tumor TNM staging. Among these factors, lymph node metastasis (LNM) status is one of the key determinants of disease staging and therapeutic strategy formulation. According to the 8th edition of the TNM classification by the International Association for the Study of Lung Cancer(IASLC) [2], the 5-year survival rate of patients with mediastinal lymph node metastasis (N2/N3 stage) is significantly lower than that of patients without lymph node involvement (N0/N1 stage). For resectable LNM (+) NSCLC patients, primary tumor resection combined with systematic lymph node dissection can improve survival outcomes [3]. However, in LNM-negative patients, this procedure not only fails to provide survival benefits but also increases the risk [4] of surgical complications and may disrupt regional immune homeostasis, thereby impairing antitumor immune surveillance.

The nuclear antigen Ki-67, a cell cycle-dependent protein, has been widely recognized as a critical biomarker for assessing proliferative activity in malignant tumors. In NSCLC, Ki-67 expression levels show a significant positive correlation [5] with tumor aggressiveness and prognosis. Studies have demonstrated that NSCLC patients with a Ki-67 labeling index ≥ 20% exhibit higher rates [6] of early recurrence and distant metastasis.

Current clinical lymph node staging primarily relies on CT morphological assessment and ¹⁸F-FDG PET/CT metabolic imaging. However, both modalities have limitations—CT offers limited diagnostic accuracy, while PET/CT suffers from insufficient specificity [7]. PET/MRI overcomes the limitations of CT’s reliance on morphological assessment alone by integrating high-resolution MRI with multifunctional imaging sequences, enabling precise identification of small metastatic lymph nodes. Simultaneously, the technique effectively discriminates between inflammatory and metastatic lymph nodes - a challenge for PET/CT - through cross-validation of metabolic parameters with diffusion and perfusion characteristics, significantly reducing false-positive rates. Studies [8, 9] have shown that PET/MRI demonstrates potential superior performance over PET/CT in NSCLC staging, and IVIM-DWI [10] technology can effectively differentiate lymph node characteristics and dynamically monitor metastatic progression through multiple quantitative parameters.

In the evaluation of the Ki-67 proliferation index, traditional biopsy [11] methods are invasive and prone to sampling errors, driving noninvasive imaging prediction as a research priority. Although studies [12, 13] have confirmed correlations between SUVmax, DWI parameters, and Ki-67 expression, single-modality imaging has inherent limitations, highlighting the urgent need for multimodal integration. IVIM technology employs a biexponential model to simultaneously quantify water molecule diffusion and microcirculatory perfusion characteristics. Its parameters—D, DDC, f, and D*—reflect the degree of restricted diffusion, microvessel density, and blood flow velocity, respectively. Existing evidence [14, 15] indicates these parameters are closely associated with Ki-67 expression in lung adenocarcinoma and lymph node metastasis risk in NSCLC. Integrated PET/MR combines the advantages of metabolic and functional imaging, enabling synchronous multidimensional assessment of tumor characteristics.

This study aims to systematically evaluate the clinical utility of integrated ¹⁸F-FDG PET/MRI in determining mediastinal LNM status and predicting the Ki-67 proliferation index in NSCLC. By quantitatively comparing the diagnostic performance of PET metabolic parameters and IVIM functional parameters and constructing an optimal multivariate logistic regression prediction model, we seek to establish a radiomics-based precision staging system for NSCLC. The findings are expected to provide objective imaging biomarkers for guiding individualized treatment strategies and prognostic assessment.

Materials and methods

Patients

This prospective study was approved by Henan Provincial People’s Hospital Ethics Review Board (NO.2021148). From July 2020 to July 2024, a total of 128 patients with suspected lung tumors on CT who underwent integrated PET/MRI were prospectively and consecutively enrolled according to the following inclusion criteria:1) maximum diameter of lung lesion ≥ 1.0 cm; 2) no prior malignancy; 3) no contraindications to MRI; 4) Pathologically confirmed NSCLC; Exclusion criteria were as follows: (1) Previous receipt of any form of antitumor therapy (n = 8); (2) Incomplete imaging data or image quality failing to meet diagnostic requirements (n = 7); (3) Missing clinical follow-up data (n = 10); (4) Absence of Ki-67 immunohistochemical testing (n = 11) (Fig. 1). All enrolled patients provided written informed consent. After screening, 92 patients with pathologically confirmed NSCLC were ultimately included in this study. Demographic and clinical characteristics including age, sex, smoking status, histological type, tumor stage, and maximum tumor diameter were recorded (Table 1).

Fig. 1.

Patient selection flowchart

Table 1.

Patient clinicopathological characteristics

Clinical characteristics			Value
Age(years)			59 ± 9
Sex		Male	43
		Female	49
Smoke		Yes	42
		No	50
Histological type		SCC	14
		AC	78
LNM		+	39
		-	53
Maximum diameter (cm)			2.7 ± 1.2
IASLC stage (2024), n (%)
Early stage (I-IIB)			64 (69.56)
Advanced stage (IIIA-IV)			28 (30.43)
Ki-67, n (%)
≤ 25%			52 (56.52)
>25%			40 (43.48)

SCC, squamous cell carcinoma; AC, adenocarcinoma; LNM, lymph node metastasis

PET/MRI scanning and image acquisition

All imaging examinations were performed using an integrated 3.0 T PET/MRI system (uPMR 790, United Imaging, Shanghai, China) equipped with a 12-channel phased-array body coil. Standardized pre-examination protocols were strictly followed: subjects were instructed to avoid strenuous exercise for 24 h prior to the examination, maintain a fasting state for at least 6 h, and confirm fasting blood glucose levels < 8.0 mmol/L. During the examination, patients were guided to maintain steady breathing patterns to minimize respiratory motion artifacts. The 18 F-FDG tracer was administered intravenously at a standard dose of 0.11 mCi/kg (4.07 MBq/kg), followed by a 40–60 min resting period prior to image acquisition. PET data acquisition was performed in the supine position with head-first orientation, covering a scan range from the lung apex to the diaphragm dome for a duration of 27 min, during which respiratory motion was monitored using an abdominal breathing belt. MRI image acquisition was conducted simultaneously with PET scanning. Attenuation correction for gamma rays was performed using a Dixon water-fat separation technique with three-dimensional T1-weighted gradient echo sequences. Image reconstruction employed the ordered subsets expectation maximization (OSEM) method. The MRI sequences included: MR-base attenuation correction (MRAC), axial T1-weighted imaging(T1WI), axial T2-weighted imaging(T2WI), and multi-b-value DWI, with specific parameters detailed in Table 2.

Table 2.

Parameters used for MR-IVIM imaging

Parameters	T1WI	T2WI	DWI	IVIM
Sequence	2D-FSE	2D-FSE	2D-SS-EPI	0.652
Orientation	Axial	Axial	Axial	Axial
TR/TE(ms)	5.06/2.1	3,315/87.8	1,620/69.6	1,620/69.6
FOV (cm2)	35 × 50	35 × 50	35 × 50	35 × 50
Matrix	303 × 456	264 × 480	202 × 256	202 × 256
Bandwidth(Hz/pixel)	260	260	2370	2370
Slice thickness(mm)	5	5	5	5
Interval (mm)	1	1	1	1
NEX	2	2	1.8	1, 1, 2, 2, 4, 4, 6, 6, 8, 10
b-values (s/mm2)	/	/	0.800	0, 25, 50, 100, 150, 200, 400, 600, 800, 1000
Respiratory compensation	Yes	Yes	Yes	Yes
Scan time	14s	2 min 26s	2 min 58 s	3 min 38 s

T1WI = T1-weighted imaging; T2WI = T2-weighted imaging; DWI = diffusion-weighted imaging IVIM = intravoxel incoherent motion FOV = field of view; TR/TE = repetition time/echo time NEX = number of excitations

Image processing and analysis

All imaging data were transferred to the uWS-MR post-processing workstation (United Imaging Healthcare, Shanghai, China) for standardized quantitative analysis. Two radiologists (M.N. and F.F.F., with 8 and 13 years of experience in thoracic oncologic imaging diagnosis, respectively) independently delineated ROIs encompassing the entire solid tumor component, while being completely blinded to the clinical data and pathological results. For PET image analysis, tumor metabolic active regions were automatically delineated using a 40% SUVmax threshold to generate volumes of interest (VOIs) Simultaneously calculated parameters included TLG, MTV, and SUVmax. Using fat-suppressed T2WI and ultrashort echo time (UTE) images as reference, ROIs were manually delineated slice-by-slice along the inner margins of solid tumor areas on ADC colored map, while excluding interfering regions such as necrosis, cystic changes, hemorrhage, gas, and calcifications. This ensured coverage of the tumor’s largest cross-sectional area while avoiding adjacent normal tissues. The software automatically reproduced the ROIs to parametric color maps (D, D*, f, and DDC) and calculated their mean values (Figs. 2 and 3). All IVIM parameter measurements were performed based on the entire lesion volume.

Fig. 2.

Multiparametric PET-MRI images and pathological images of different patients: (a-h) patient with adenocarcinoma (white arrow, LNM-positive).(i-p) patient with squamous cell carcinoma (white arrow, LNM-negative).(a, i) T2-weighted images (T2WI); (b, j) ¹⁸F-FDG PET and UTE fusion images; (c, k) Pseudocolored ADC maps; (d, l) Pseudocolored D maps; (e, m) Pseudocolored DDC maps; (f, n) Pseudocolored D* maps; (g, o) Pseudocolored f maps; (h, p) Pathological images (×200)

Fig. 3.

The performance of various parameters and a combined model (MTV + DDC) was evaluated using ROC curves. The combination model achieved the highest AUC (0.821) in both analyses. (a) The AUCs of SUVmax, MTV, and TLG were 0.629, 0.754, and 0.726, respectively. (b) The AUCs of ADC, D, and DDC were 0.685, 0.656, and 0.736, respectively

The formula used to determine the IVIM sequence parameters is presented below. Sb = S0×[(1-f) × exp(-bD) + f × exp (-b ×D*)] for the bi-exponential IVIM model illustrates the relationship between the DWI signal intensity and the b factor. Sb represents the signal intensity, and b stands for the sensitivity factor. The [16]diffusion coefficient is represented by D, and the perfusion fraction by f. D* is used to represent the pseudo-diffusion factor.

Stretched exponential model: Sb = S0×exp[-(b×DDC)α], the parameters for the biexponential and stretched exponential models were fitted using ten different b-values (0, 25, 50, 100, 150, 200, 400, 600, 800, and 1000 s/mm²), where DDC represents the distributed diffusion coefficient and α denotes the water molecular diffusion heterogeneity index.

Pathological diagnosis process and standards

The surgically resected lymph node specimens were processed through dissection, sectioning, embedding, and histological staining (hematoxylin and eosin) before being analyzed by pathologists. Ki-67 expression was assessed via immunohistochemistry using mouse anti-human Ki-67 monoclonal antibodies. Ki-67 positivity was defined as nuclear staining appearing brownish-yellow or brown. A senior pathologist, blinded to the patient’s clinical and imaging data, evaluated the lymph node metastasis status and Ki-67 expression.

Statistical analysis

Data analysis was performed using MedCalc (version 22.0), R (version 4.2.1), and SPSS (version 29.0) software. The intraclass correlation coefficient (ICC) was employed to assess intra- and inter-observer agreement, with interpretation criteria as follows: 0.75–0.90 indicated good agreement, and > 0.90 indicated excellent agreement. The normality of each parameter was evaluated using the Kolmogorov-Smirnov test. Normally distributed variables were expressed as mean ± standard deviation (Mean ± SD) and compared between groups using Student’s t-test. Non-normally distributed variables were expressed as median (interquartile range) and compared using the Mann-Whitney U test. ROC curves were plotted, and the area under the ROC curve (AUC) was calculated to describe diagnostic performance. The optimal threshold was determined based on the maximum Youden index. The DeLong test was used to compare differences in AUC between parameters (individually or in combination). Logistic regression analysis was performed to identify the optimal predictive model. Spearman’s rank correlation coefficient was used to evaluate the correlation between PET/IVIM parameters and Ki-67 expression. The significance level was set at 0.05.

Results

Consistency testing

Measurements from both observers demonstrated excellent agreement across all parameters. The intraclass correlation coefficient (ICC) values for ADC, D, D*, f, DDC, SUVmax, MTV, and TLG were all greater than 0.80.

Comparison of parameters between groups

The LNM (+) group showed significantly higher SUVmax, MTV, and TLG compared to the LNM (-) group. Conversely, ADC, D, and DDC were significantly lower in the LNM (+) group. Subsequently, an FDR correction applied to these significant parameters demonstrated that all of them remained statistically significant (Table 3).

Table 3.

Comparison of ¹⁸F-FDG PET and IVIM parameters for LNM status

Variables	LNM-positive	LNM-negative	P-value	FDR-corrected P-value
N	39	53	-	-
SUVmax(g/cm3)	6.87 (3.90, 8.84)	4.65 (3.07, 7.43)	0.035	0.035
MTV(cm3)	9.11 (6.27, 28.48)	4.07 (2.47, 9.87)	< 0.001	0.002
TLG(g)	29.49 (14.35, 80.17)	12.54 (3.45, 22.49)	< 0.001	0.002
ADC(×10− 3mm2/s)	0.78 (0.55, 1.20)	1.10 (0.85, 1.36)	0.003	0.005
D(×10− 3mm2/s)	0.70 (0.37, 0.98)	0.89 (0.69, 1.12)	0.011	0.013
DDC×10− 3mm2/s)	2.43 (1.83, 3.43)	4.73 (2.47, 6.35)	< 0.001	0.002
D*(×10− 3mm2/s)	71.49 (34.83, 105.42)	69.58 (45.77, 100.59)	0.965	-
f (%)	41.05 (31.93, 55.34)	42.38 (26.54, 51.80)	0.690	-

SUVmax = maximum standardized uptake value; MTV = metabolic tumor volume; TLG = total lesion glycolysis; ADC = apparent diffusion coefficient; D = true diffusion coefficient; DDC = distributed diffusion coefficient; D* = pseudo diffusion coefficient; f = perfusion fraction; LNM = lymph node metastasis

Regression analyses

In the identification of LNM (+) and LNM (-), univariate (Table 4) analysis showed that age, gender, histology, SUVmax, MTV, TLG, ADC, D, and DDC were predictors, and multivariate analysis revealed that only MTV and DDC were independent predictors (Fig. 4).

Table 4.

Univariate and multivariate analyses

Parameters	VIF	Univariate analyses		Multivariate analyses
		OR (95% CI)	P-Value	OR (95% CI)	P-Value
Age	1.099	0.969 (0.926–1.014)	0.180	-	-
Gender	1.293	0.960 (0.419–2.199)	0.923	-	-
Histology	1.422	0.185 (0.039–0.880)	0.034	0.147 (0.020–1.062)	0.057
SUVmax	2.467	1.145 (1.007–1.303)	0.039	0.883 (0.689–1.132)	0.326
MTV	2.359	1.119 (1.047–1.196)	< 0.001	1.128 (1.017–1.251)	0.023
TLG	3.053	1.017 (1.005–1.029)	0.007	1.006 (0.988–1.024)	0.506
ADC	3.357	0.206 (0.066–0.646)	0.007	0.305 (0.027–3.502)	0.340
D	3.235	0.231 (0.069–0.775)	0.018	0.705 (0.056–8.806)	0.786
DDC	1.498	0.618 (0.477-0.800)	< 0.001	0.673 (0.503–0.902)	0.008
D*	1.311	0.999 (0.989–1.009)	0.802	-	-
f	1.132	0.997 (0.984–1.010)	0.602	-	-

SUVmax = maximum standardized uptake value; MTV = metabolic tumor volume; TLG = total lesion glycolysis; ADC = apparent diffusion coefficient; D = diffusion coefficient; DDC = distributed diffusion coefficient; D* = pseudo diffusion coefficient; f = perfusion fraction. VIF = variance inflation factor, a VIF value < 5 indicates negligible multicollinearity. All factors with P < 0.1 in univariate analyses were included in multivariate regression analyses. OR = odds ratio; CI = confidence interval. The bold typeface in the table indicates the logistic regression analyses with statistical significance (significance level = 0.05)

Fig. 4.

(a-f) Comparison of parameters (SUVmax, MTV, TLG, ADC, D, and DDC) between the LNM-positive and LNM-negative groups is shown. * P < 0.05, *** P < 0.001

Diagnostic value of different parameters

To evaluate the predictive efficacy of different parameters for lymph node metastasis, ROC curves were generated for patients with and without lymph node metastasis. The AUC values were as follows: (MTV + DDC) > MTV > DDC > TLG > ADC > D > SUVmax (AUC = 0.821, 0.754, 0.736, 0.726, 0.685, 0.656, and 0.629). Statistical analysis revealed significant differences between the AUC of the combined model (MTV + DDC) and those of DDC, TLG, ADC, D, and SUVmax individually (Table 5). Although no statistically significant difference was observed between the combined model and MTV alone, the combined approach demonstrated improved both specificity and sensitivity.

Table 5.

ROC analysis of the diagnostic performance for predicting LNM status

Parameters	AUC (95% CI)	Youden Index	Cut-off Value	Sensitivity (%)	Specificity (%)	Comparison with Combined Diagnosis
SUVmax	0.629 (0.513–0.746)	0.3202	6.26	64.10	67.93	P = 0.0006
MTV	0.754 (0.655–0.852)	0.4039	7.16	74.36	66.03	P = 0.0998
TLG	0.726 (0.621–0.832)	0.4267	25.45	61.54	81.13	P = 0.0270
ADC	0.685 (0.568–0.802)	0.3875	0.79	53.87	84.91	P = 0.0462
D	0.656 (0.538–0.774)	0.3014	0.80	64.1	66.04	P = 0.0161
DDC	0.736 (0.633–0.838)	0.4485	3.51	76.92	67.93	P = 0.0135
MTV + DDC	0.821 (0.736–0.907)	0.5307	–	79.49	73.59	–

SUVmax = maximum standardized uptake value; MTV = metabolic tumor volume; TLG = total lesion glycolysis; ADC = apparent diffusion coefficient; D = diffusion coefficient; DDC = distributed diffusion coefficient; D* = pseudo diffusion coefficient; f = perfusion fraction

Correlation analysis

SUVmax, MTV, and TLG were positively correlated (Fig. 5)with Ki-67 (r = 0.232, P = 0.026; r = 0.213, P = 0.041; r = 0.3, P = 0.004), while ADC values and D values were negatively correlated with Ki-67 (r = − 0.327, P = 0.001; r = − 0.24, P = 0.021).

Fig. 5.

(a–f) Correlation between Ki-67 and SUVmax, MTV, TLG, ADC, D, and DDC (r = 0.232, 0.213, 0.3, − 0.327, − 0.24, and − 0.204, P = 0.026, 0.041, 0.004, 0.001, 0.021, and 0.051)

Model validation

The calibration curve based on bootstrap (1000 repetitions) internal validation shows good agreement between the combined diagnostic prediction results and the actual observed results (Fig. 6).

Fig. 6.

Calibration plot of the validation model for predicting lymph node metastasis in NSCLC

Discussion

At present, a small number of studies have applied IVIM-PET in the assessment of NSCLC. The relevant results show that the related derivative parameters can not only effectively distinguish benign and malignant solitary pulmonary nodules [16], but also contribute to the assessment of cell proliferation status [14]. Inspired by the above research, this study further explored the value of IVIM-PET-related parameters in the assessment of lymph node metastasis status of NSCLC and their correlation with Ki-67. It was found that SUVmax, MTV, TLG, ADC, D, D*, f, and DDC could all effectively differentiate LNM status in NSCLC. Among these, the combined diagnostic model of MTV + DDC exhibited the highest predictive efficacy. Furthermore, SUVmax, MTV, TLG, and D values all showed correlations with Ki-67 expression. These findings suggest that integrated PET/MRI may serve as a non-invasive biomarker for assisting in the evaluation of LNM and Ki-67 status.

ADC, D, and DDC are all diffusion-related parameters of IVIM, which can quantify the diffusion movement of water molecules. Previous studies have shown that lesions with a higher degree of malignancy, due to faster cell proliferation, have significantly restricted internal water molecule movement, and their spread rate is often lower than that of lesions with a lower degree of malignancy [10, 17]. This investigation revealed significantly diminished ADC, D, and DDC values in LNM-positive group compared to their LNM-negative counterparts (P < 0.05), and at the same time, ADC and D were all negatively correlated with Ki-67. One possible explanation is that lesions in the LNM positive group and those with higher Ki-67 expression have a higher degree of malignancy. Therefore, their cell proliferation is more vigorous, and the diffusion movement of water molecules is more significantly restricted, which in turn leads to a decrease in ADC, D, and DDC [14, 18]. Compared with ADC and D, DDC represents the contribution of non-Gaussian diffusion and tumor heterogeneity, and can reflect the average diffusion information in all directions. Therefore, it is more sensitive to changes in diffusion characteristics. A previous study [19] comparing benign and malignant lymph nodes in NSCLC found that metastatic lymph nodes had lower DDC values than non-metastatic lymph nodes, with specificity reaching up to 97%. The research by Ke W et al. [20] also demonstrated that DDC could more effectively differentiate benign from malignant solitary pulmonary lesions compared to ADC. The multifactorial logistic regression analysis in this study indicated that among diffusion parameters, only DDC is an independent predictor of NSCLC LNM, consistent with the results of the aforementioned studies, fully demonstrating that DDC has a significant advantage over ADC and D values in the assessment of NSCLC.

D* and f are parameters used by IVIM to reflect microcirculation perfusion in tissues. Theoretically, the more sufficient the microcirculation perfusion in tissues is, the greater the value of D* and f will be [21, 22]. In this study, the values of D* and f had no statistical significance in evaluating LNM status, and there was no significant correlation with the expression of Ki-67. This is similar to the study of Ye X et al. [15], suggesting that D* and f may have difficulty playing a role in the assessment of LNM and Ki-67 in NSCLC patients. We speculate that this might be related to the absence of significant differences in microcirculation perfusion among NSCLC with different lymph node metastasis states and Ki-67 expression. We speculate that this might be related to the absence of significant differences in microcirculation perfusion among NSCLC with different lymph node metastasis states and Ki-67 expression. In addition, the low reproducibility and high variability of the D* and f values might also be one of the important reasons for this result [23].

Previous studies [24, 25] have shown that the expression of glucose transporter-1(Glut-1) is associated with lymph node metastasis in various tumors, leading to increased FDG uptake and corresponding quantitative indicators. Meanwhile, studies [26, 27] have indicated that different histological subtypes and exhibit varying affinities for FDG. In this study, univariate logistic regression indicated that histology was a predictor for lymph node metastasis in NSCLC; however, it failed to emerge as an independent predictor in the multivariate analysis. This discrepancy can be primarily attributed to the skewed distribution of our study sample, specifically the overrepresentation of adenocarcinoma. Multiple studies have demonstrated that the SUVmax value of tumors is a predictive marker [28, 29] for lymph node metastasis in NSCLC, and in our study, an SUVmax ≥ 6.3 could distinguish between the presence or absence of lymph node metastasis. The research by Li et al. [30] also supports our findings, as they suggested that a primary tumor SUVmax ≥ 4.3 has value in determining LNM in NSCLC. Additionally, our study explored two parameters, MTV and TLG, based on tumor volume. Various studies [31, 32] have reported that the MTV and TLG values of primary tumors provide information about tumor volume and activity, which may more accurately assess the metabolic burden of tumors compared to SUVmax. Seong Park et al. [31] reported in a study of 139 NSCLC patients that MTV demonstrated better predictive performance than other PET parameters in evaluating lymph node status in NSCLC. In this study, multivariate logistic regression analysis revealed that MTV was the only independent metabolic predictor, further confirming its predictive value for lymph node metastasis in NSCLC. This may be related to the fact that SUVmax only reflects the metabolic level of a single voxel within the lesion while ignoring the overall changes in the tumor. Moreover, differences between studies, including variations in the study population, injected radioactivity, image reconstruction parameters, and SUVmax measurement methods, may be important factors contributing to the inconsistencies between our study and previous research. In addition, our research also found that SUVmax, MTV and TLG are all positively correlated. One possible explanation is that lesion cells with higher Ki-67 proliferate more vigorously and have more significant energy metabolism, thereby causing an increase in SUVmax, MTV and TLG.

This study utilized multivariate logistic regression analysis to establish a combined prediction model for lymph node metastasis in NSCLC patients, finding that the combined model of MTV + DDC demonstrated superior predictive performance compared to any single model alone. In previous studies, lymph nodes with FDG uptake higher than the metabolic level of background lymph nodes were considered diagnostic criteria for lymph node metastasis. However, inflammatory or granulomatous lesions can also lead to increased uptake. This study confirmed the ability of such criteria to predict lymph node metastasis while also demonstrating its relatively low sensitivity. In contrast, the MTV + DDC combined prediction model exhibited better predictive value (AUC 0.821, sensitivity 79.49%, specificity 73.59%). By integrating metabolic burden and microenvironment heterogeneity information, the MTV + DDC combined model overcomes the biological limitations of single imaging parameters, providing a more reliable predictive tool for lymph node metastasis in NSCLC.

As a key biomarker for assessing tumor cell proliferative activity, Ki-67 demonstrated modest correlations with multiple metabolic parameters (SUVmax, MTV, TLG) and diffusion parameters (ADC, D) in this study. The underlying mechanism involves elevated Ki-67 expression indicating active tumor proliferation, which simultaneously increases metabolic demand and cellular density, consequently leading to enhanced FDG uptake and restricted water molecule diffusion. These findings highlight the potential of multiparametric imaging models [12, 14] for noninvasive Ki-67 evaluation. Integrating PET metabolic parameters with IVIM diffusion parameters enables comprehensive characterization of tumor biological features. This multimodal imaging fusion strategy effectively addresses tumor heterogeneity and biopsy sampling limitations, providing a novel approach for preoperative noninvasive assessment of Ki-67 expression. Notably, this study found no statistically significant correlation between the Ki-67 index and perfusion-related parameters (D* and f values). Apart from the two aforementioned lung cancer-related studies, Wang et al. [33] also reached a similar conclusion in a bladder cancer study, stating that restricted water molecule diffusion effectively reflects cell proliferation levels, whereas blood flow perfusion parameters are susceptible to multiple interfering factors. This may be because benign conditions such as tissue hyperemia and edema can also alter blood perfusion, indicating that D* and f values are influenced by too many confounding factors and do not accurately reflect tumor proliferative status.

This study has several limitations. First, our research adopted a single-center, single-scanner design, which may lead to limitations in technical reproducibility and potential case selection bias. Second, the predictive model was based on the primary tumor rather than lymph nodes, imposing constraints in identifying the location and number of metastatic lymph nodes. Third, PET/MR images for each patient were acquired under free-breathing conditions, introducing respiratory motion artifacts into the images. Fourth, lesions with a maximum diameter smaller than 1 cm were excluded from this study due to noise interference, which may have introduced a certain degree of selection bias. Fifth, the Dixon sequence-based MR attenuation correction method has known limitations and potential biases in pulmonary tissue. Sixth, the lack of adjustment for potential confounders such as tumor size or stage weakens the biological interpretation. Seventh, the study team did not include a nuclear medicine physician, which could potentially impact the assessment of PET quantitative metrics and lymph node interpretation. Finally, the absence of external validation for the predictive model, coupled with the weak correlation between certain indicators and Ki-67, collectively constrains the external validity of the study conclusion.

Conclusions

Multiparametric PET/MRI can be used to evaluate lymph node metastasis and Ki-67 proliferation status in non-small cell lung cancer (NSCLC). The combined predictive model of MTV and DDC provides a novel clinical approach for assessing lymph node metastasis status. Furthermore, the observed correlations between certain imaging parameters and Ki-67 expression, combined with this multimodal methodology, allow for preoperative identification of high-risk metastatic regions. This capability assists clinicians in performing targeted biopsies of central or mediastinal lymph nodes, avoiding unnecessary extended dissections, while facilitating early planning of neoadjuvant therapies or surgical strategy modifications, ultimately supporting the development of personalized treatment strategies for NSCLC patients.

Author contributions

QQC, WW, and NM conceptualized the ideas, designed the study, analyzed the data, constructed the models, and authored the manuscript text. XYW, YL, JWZ and JYP were responsible for recruiting and scanning the patients. YPW, ZH, YY and ZW assisted in the statistical analysis and model building. QYL and FFF critically revised the manuscript for important intellectual content. FFF and MYW provided administrative, technical, or material support and supervised the study. All authors reviewed the manuscript. All authors read and approved the final manuscript.

Funding

The National Key R&D Program of China (2023YFC2414200), the National Natural Science Foundation of China (82441022, 82441023, 82371934), the Special Training Program for Clinical Research Physicians under the “Three 100” lnitiative of Henan Province (HNCRD202406), the Natural Science Foundation of Henan Province (252300420565), the Science and Technology Project of Henan Province (252102311105), the Key Project of Henan Province Medical Science and Technology Project’s (LHGJ20210001, LHGJ20240036, LHGJ20240053).

Data availability

Due to privacy restrictions, raw data cannot be made available free of charge, but datasets used and/or analyzed during the current study may be obtained from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

All procedures performed in studies involving human participants were in accordance with the ethical standards of the Institutional Research Committee of Henan Provincial People’s Hospital and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. This prospective study was approved by Henan Provincial People’s Hospital Ethics Review Board (NO.2021148). All participants provided written informed consent to participate in this study. All data is analyzed anonymously.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.