Imaging biomarkers related to tumor-associated macrophage in immunotherapy treatment planning for non-small cell lung cancer

Introduction

Immune checkpoint inhibitors (ICIs) have dramatically improved the prognosis of lung cancer, yet only a modest proportion of patients achieve durable responses and derive clinical benefit (20–50% in non-small cell lung cancer (NSCLC) (1,2). A subset of patients may even experience hyperprogression (4–29% in NSCLC) (3-9) or pseudoprogression—atypical responses that are difficult to discriminate within the first 8 weeks (10). The heterogeneous responses and variable clinical outcomes pose significant challenges in managing ICI-treated patients. Currently, most approved or investigational predictive biomarkers, such as programmed cell death ligand 1 (PD-L1) status and tumor mutational burden (TMB), primarily focus on the primary tumor. However, these markers are insufficient for accurately identifying patients who will truly benefit from immunotherapy (11-13), particularly in distinguishing hyperprogression and pseudoprogression (5,7). Lymph nodes harbor a diverse population of immune cells that contribute to the body’s anti-tumor immune response, and recent studies have demonstrated that interactions between programmed cell death 1 (PD-1) and PD-L1 in tumor-draining lymph nodes are directly associated with melanoma prognosis (14). Thus, the supplementary predictive value of lymph nodes in ICI therapy warrants further investigation.

Current lymph node assessment relies on invasive biopsy or surgical procedures. Conversely, effective preservation of draining lymph nodes for a specific duration may enhance ICI treatment efficacy (15). Therefore, a non-invasive method to analyze lymph nodes is urgently needed. Drawing inspiration from the widely used radiomics analysis of primary tumors and peritumoral regions—which enables quantitative characterization of lesion features and shows strong associations with immunotherapy outcomes in lung cancer (16-23)—applying radiomics analysis to lymph nodes may provide valuable insights. This suggests that quantitative indices derived from lymph nodes could offer supplementary information for predicting diverse clinical outcomes of ICI treatment.

Additionally, compelling evidence indicates that tumor-associated macrophages (TAMs) promote processes such as angiogenesis, lymphangiogenesis, tumor growth, and progression in solid tumors (24). M1-polarized and M2-polarized TAMs coexist in the tumor microenvironment. In lung adenocarcinoma, a greater number of M2-polarized TAMs have been detected compared to M1-polarized TAMs, and they correlate more strongly with lymph node metastasis than M1-polarized TAMs (25).

Here, we investigated the supplementary value of baseline lymph node radiomics features in predicting diverse clinical outcomes, including durable clinical benefit (DCB), best overall response (BOR), progression-free survival (PFS), overall survival (OS), and pseudoprogression, in patients with advanced NSCLC. We further explored the biological significance of the radiomics model associated with TAMs to assist clinicians in making informed clinical decisions. We present this article in accordance with the TRIPOD reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-323/rc).

Methods

Data and study design

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study received ethical approval from the Institutional Review Board of Shandong Cancer Hospital and Institute, Shandong Provincial Hospital Affiliated with Shandong First Medical University (No. SDTHEC2020001015), with informed consent waived owing to the retrospective design. The inclusion criteria were as follows: (I) patients with pathologically confirmed NSCLC receiving anti-PD-(L)1 immunotherapy; (II) contrast-enhanced computed tomography (CT) scans obtained within two weeks prior to immunotherapy initiation serving as baseline scans; (III) a minimum follow-up period of 6 months from immunotherapy initiation without intolerable treatment-related toxicity; (IV) patients with measurable primary lesions (longest diameter ≥10 mm), with or without lymph nodes (short axis ≥10 mm) (26). Patients who failed to meet these criteria were excluded. Ultimately, a total of 212 patients (157 males, 55 females) were included in the final cohort, which was then divided into three groups: a training cohort (n=105; 80 males, 25 females), a test cohort (n=70; 51 males, 19 females), and an external test cohort (n=37; 26 males, 11 females). The patient enrollment and grouping flowchart is presented in https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf.

In this study, the primary endpoint was DCB versus no durable benefit (NDB), defined using a binary cutoff of PFS at 6 months (27,28). Secondary endpoints included PFS and OS. PFS was defined as the time from the initiation of ICI therapy to the occurrence of intolerable treatment-related toxicity, disease progression, or death, determined using the Immune Response Evaluation Criteria in Solid Tumors (iRECIST) (27). OS was defined as the time from the first ICI administration to death from any cause. Additionally, BOR, pseudoprogression, and hyperprogression were recorded from treatment initiation to disease progression. BOR was defined as the optimal response recorded between the date of the first dose and the earlier of either the date of initial confirmed tumor progression (per objective assessment) or the date of subsequent therapy, whichever occurred first (1). Tumor progression was defined using the Response Evaluation Criteria in Solid Tumors (RECIST 1.1) (26). Hyperprogression was defined as time-to-treatment failure (TTF) <2 months, >50% increase in tumor burden, and >2-fold acceleration in progression rate compared with pre-immunotherapy imaging (29), whereas pseudoprogression was defined as an initial increase in primary tumor size or new lesion appearance followed by subsequent tumor regression (30). The study schema is illustrated in Figure 1.

Figure 1 Schematic workflow of this study. CT, computed tomography; DCB, durable clinical benefit; ICI, immune checkpoint inhibitor; IHC, immunohistochemistry; LASSO, least absolute shrinkage and selection operator; OS, overall survival; PD-L1, programmed cell death ligand 1; PFS, progression-free survival; PR/CR, partial/complete response; ROI, region of interest; RS, radiomic signature.

CT acquisition and lymph node selection

CT scans were performed using a 128-row CT scanner (MX 8000, Philips Medical Systems, Amstelveen, Netherland) following standardized protocols, with the following parameters: 1.25 mm collimation, 1.375 pitch, 120-kV tube voltage, 200-mA tube current, 512×512 matrix, and 5.0 mm slice thickness. All reconstructed CT images were generated using a medium-sharp reconstruction algorithm, with a slice thickness of 1 mm and 1 mm increment. For each patient, contrast-enhanced CT scans were acquired via intravenous contrast injection (2.5 mL/s; 1.5 mL/kg), with images obtained at 60 seconds post-injection used for analysis. CT images were interpreted and contoured by two radiologists (each with over 15 years of experience) from their respective hospitals. One radiologist initially performed the interpretation and contouring, followed by a review and potential revisions by the second radiologist as needed. Lymph nodes selected for feature extraction were those with the largest size in their respective zones (short axis ≥10 mm).

Development of the radiomics signature

Based on the CT images and masks, three types of radiomic features were automatically extracted as initial features by PyRadiomics toolkit: (I) whole tumor features, (II) peritumoral features focusing on 5 and 10 mm margins from the tumor boundary (Peri5mm, Peri10mm), (III) lymph node features. For patients with multiple lymph nodes, the average values of all lymph node features were calculated as the final representation; for patients without detectable lymph nodes, all lymph node-related features were assigned zero values. Finally, 1,134 features, including whole tumor features, peritumoral features focusing on 5 and 10 mm margins, and lymph node features, were scaled into the range [0,1] with unity-based normalization, as shown in https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf. Feature extraction was performed as shown in https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf.

Given the redundancy of the extracted features, all the features were grouped based on consensus clustering (https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf), and only the features with the largest receiver operating characteristic curves (AUC) in each group were left to reduce the dimensionality. Then, the least absolute shrinkage and selection operator (LASSO) Cox regression method was further used to select the most useful predictive features with nonzero coefficients and generate the radiomics signatures through a linear combination weighted by the corresponding coefficients. 10-fold cross-validation was used to determine the penalty parameter in LASSO by minimum mean cross-validated error. This pipeline yielded four key radiomic signatures: RS_WT (whole tumor), RS_Peri5mm (5 mm peritumoral), RS_Peri10mm (10 mm peritumoral), and RS_Lymph (lymph node).

In order to investigate the added value of the features from peritumor and lymph node, generalized linear model (GLM) using forward stepwise selection algorithm was performed to generate combined signatures RS_WP (combination of whole tumor and peritumor signatures) and RS_WPL (combination of whole tumor, peritumor and lymph node signatures).

Immunohistochemistry

All specimens obtained from primary tumors via needle biopsy prior to ICI treatment were included in the immunohistochemical analysis. Immunohistochemistry was performed on formalin-fixed, paraffin-embedded (FFPE) human tissue sections using standard antigen retrieval protocols in citrate buffer. Expressions of TAMs, M1-like TAM, and M2-like TAM were detected using CD68, CD86, and CD163, respectively, as detailed in the https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf. All markers were scored based on the percentage of immunoreactive cells relative to the total number of cells in the tissue. Patient samples were categorized into two groups (DCB and NDB) according to RECIST criteria.

Statistical analysis

The Wilcoxon rank-sum test and Jonckheere-Terpstra test were used to assess differences in continuous variables, while the chi-square test and Fisher’s exact test were applied for categorical variables. The AUC, accuracy, sensitivity, specificity, and the 95% confidence interval (CI) by the Delong method (29) were used to assess the classification ability. The cut-off was established using the maximum Youden index (i.e., specificity + sensitivity −1) in the training cohort for different treatment type. The AUC, sensitivity, and specificity of different models were compared using the Delong test and the McNemar Test. The Kaplan-Meier survival analysis for PFS and OS was performed, with the log-rank test used to compare survival curves. Cox regression analysis was applied to measure the association between the model’s predicted scores and patients’ DCB and PFS. Pearson correlation and multiple linear regression analysis (Stepwise Feature Selection Method: Forward) were used to examine the correlation between model and biomarkers. All statistical analyses were implemented on R 3.5.2 and SPSS 26.0. A two-sided P value less than 0.05 was regarded as statistically significant.

Results

Clinical characteristics

Between January 1, 2017, and October 1, 2019, a total of 175 patients from Shandong Cancer Hospital and Institute, along with 37 patients from Zhongshan Hospital (Affiliated with Fudan University), were enrolled in this study. Key baseline characteristics are summarized in Table 1. Participants were randomly assigned to training (n=105), test (n=70), and external test (n=37) cohorts, with no significant differences in demographic characteristics across the three groups. The prevalence of DCB was 55.2% in the training cohort, 57.1% in the test cohort, and 35.1% in the external test cohort. A total of 858 lymph nodes were analyzed, including 458 from the training cohort, 277 from the test cohort, and 123 from the external test cohort. Distributions and comparisons of lymph nodes between DCB and NDB groups are detailed in https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf. There was no significant difference in the number of lymph nodes of different sites between DCB and NDB patients in three cohorts (Chi-squared test, P>0.05). Median PFS was 6.83 months in the training cohort, 11.57 months in the test cohort, and 9.30 months in the external test cohort. Corresponding rates of progressive disease (PD) were 24.8% (26/105), 22.9% (16/70), and 29.7% (11/37) in these groups, respectively. Median OS was 21.17 months in the training cohort, 22.53 months in the test cohort, and 17.33 months in the external test cohort, with mortality rates of 57.1% (60/105), 51.4% (36/70), and 48.6% (18/37), respectively. Representative examples of immunohistochemical results from four patients are shown in Figure 2.

Figure 2 Examples of four different patients with their immunohistochemistry photomicrograph (original magnification, ×200). (A) Patient without lymph node had DCB. (B) Patient without lymph node had NDB. (C) Patient with lymph node had DCB. (D) Patient with lymph node had NDB. DCB, durable clinical benefit; NDB, no durable benefit; PFS, progression-free survival; RS, radiomic signature; WPL, whole tumor, peritumor and lymph node

Development of the radiomics signature

After consensus clustering analysis to eliminate redundant features, 15 uncorrelated features (four whole tumor features, four peritumoral-5 mm features, three peritumoral-10 mm features and four lymph node features) were remained. Trough LASSO and GLM method, six radiomics signatures were generated as shown in https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf.

Improved predictive value of the lymph node features in DCB prediction

The prediction results of the different radiomics signatures are shown in Table 2 and Figure 3A-3C. The RS_WPL achieved the highest AUC values of 0.84 (95% CI: 0.76–0.90), 0.82 (95% CI: 0.64–0.90), and 0.81 (95% CI: 0.67–0.93) in the training, test, and external test cohorts, respectively. Except for the RS_Peri10mm, which had the lowest AUC (0.59–0.64), the RS_WT, RS_Peri5mm, RS_Lymph, and RS_WP exhibited comparable AUC values (0.61–0.75).

Figure 3 Quantitative metrics of different radiomics signatures. (A-C) ROC curves of different radiomics signatures of the training, test and validated cohort, respectively. (D) The sensitivity of the different radiomics signatures on the training, test and external test cohorts. ***, P<0.001; *, P<0.05. Compared to other radiomics features, the sensitivity of WPL was significantly improved. In the training cohort, the P values compared to RS_WT, RS_Peri5mm, RS_Peri10mm, RS_lymphnode, and RS_WP were 0.04, 0.01, 0.01, 0.03, and 0.04, respectively. In the test cohort, significant differences were observed compared to RS_Peri, RS_ Peri 10mm, and RS_lymphnode, with P values of 0.03, 0.043, and 0.007. In the external test cohort, significant differences were found compared to RS_Peri5mm and RS_ Peri10mm, with P values of 0.03 and 0.02. (E) The specificity of the different radiomics signatures on the training, test and external test cohorts. (F) Comparison between radiomics signature and PD-L1 status. (G,H) The distribution of RS_WPL in PD-L1 negative and PD-L1 positive patients, respectively. AUC, area under the receiver operating characteristic curve; PD-L1, programmed cell death 1-ligand 1; ROC, receiver operating characteristic; RS, radiomic signature; WPL, whole tumor, peritumor and lymph node.

With the incorporation of lymph node features, the sensitivity of RS_WPL was increased to 0.84 (0.74–0.93), 0.85 (0.72–0.95), and 0.92 (0.77–1.00) in the three cohorts, respectively, which was significantly higher than that of the other radiomics signatures (P<0.05 or P<0.01), while specificity remained comparable (Figure 3D,3E).

Among the 84 patients in the PD-L1 cohort, the AUC values for RS_WP, RS_WPL, and the PD-L1-based signature (RS_PD-L1) in predicting DCB were 0.74, 0.83, and 0.59, respectively (Figure 3F). Figure 3G,3H show the distribution of RS_WPL in PD-L1-negative and PD-L1-positive patients. Among 28 PD-L1-negative patients, eight were identified as DCB and thus should receive immunotherapy. In contrast, 15 of 56 PD-L1-positive patients did not benefit from immunotherapy and should not receive it. These findings suggest that RS_WPL could complement PD-L1 in identifying patients who would benefit from immunotherapy.

Prognostic performance of the radiomics signature in PFS and OS

Through Kaplan-Meier survival analysis, RS_WPL was able to significantly predict PFS and OS in training (log-rank, P<0.01, P=0.002), test (log-rank, P=0.01, P=0.02), and external test cohorts (log-rank, P=0.02, P=0.25), respectively. Applying 0.15 as the cutoff, patients were grouped in high RS_WPL and low RS_WPL. Patients with high RS_WPL had longer PFS and OS (Figure 4A-4F).

Figure 4 Performance of the RS_WPL in prognosis prediction. (A-C) The PFS of the patients relative to the RS_WPL (low versus high) in the training, test and validated cohorts, respectively. (G,H) The PFS of the patients with positive and negative PD-L1 status relative to the RS_WPL (low versus high) in the PD-L1 cohorts, respectively. (D-F) The OS of the patients relative to the RS_WPL (low versus high) in the training, test and external test cohorts, respectively. (I,J) The OS of the patients with positive and negative PD-L1 status relative to the RS_WPL (low versus high) in the PD-L1 cohorts, respectively. CI, confidence interval; HR, hazard ratio; OS, overall survival; PD-L1, programmed cell death ligand 1; PFS, progression-free survival; RS, radiomic signature; WPL, whole tumor, peritumor and lymph node.

To further evaluate the predictive performance of RS_WPL across PD-L1 subgroups—given the well-established association between PD-L1 status and survival outcomes—stratified analyses were performed. Results (Figure 4G-4J) demonstrated that high RS_WPL remained significantly associated with prolonged PFS and OS in PD-L1-positive patients (HR for PFS =0.36, 95% CI: 0.17–0.76, P=0.002; HR for OS =0.45, 95% CI: 0.21–0.97, P=0.02). In contrast, among PD-L1-negative patients, there were no significant differences in PFS or OS between high- and low-RS_WPL groups (HR for PFS =0.07, P=0.07; HR for OS =0.57, P=0.28). These findings suggest that RS_WPL may enhance the prediction of PFS and OS specifically in PD-L1-positive patients.

Predictive value of RS_WPL in typical and atypical response

We further evaluated the performance of RS_WPL in predicting typical (PR/CR, SD, PD) and atypical treatment responses (pseudoprogression and hyperprogression), finding the results satisfactory (Figure 5). The RS_WPL was predictive of typical response in training (Jonckheere-Terpstra test, P=0.002) and test cohorts (Jonckheere-Terpstra test, P=0.001). However, there was a difference trend in the validate cohort (Jonckheere-Terpstra test, P=0.07). The AUC values were 0.66 (95% CI: 0.57–0.79), 0.71 (95% CI: 0.58–0.88) and 0.71 (95% CI: 0.42–0.87). Pseudoprogression and hyperprogression had statistically different distribution of RS_WPL (Wilcoxon rank-sum test, P=0.03) with the high AUC [0.80 (95% CI: 0.63–0.92)]. These findings suggest that RS_WPL may potentially identify pseudoprogression, thereby assisting oncologists in adjusting treatment plans in a timely manner.

Figure 5 Performance of the RS_WPL in discriminating other clinical outcome. (A-C) The box plots of the distribution of RS_WPL of the training, test and external test cohorts according to different best response, respectively. (D) ROC curves of RS_WPL in PR/CR identification in the training, test and external test cohorts. (E) The comparison of the distribution of RS_WPL between pseudoprogression and hyper progression. (F) ROC curves of RS_WPL in pseudoprogression identification. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CR, complete response; PD, progressive disease; PR, partial response; ROC, receiver operating characteristic; RS, radiomic signature; SD, stable disease; WPL, whole tumor, peritumor and lymph node.

Physiological significance of the radiomics signatures

Baseline characteristics of patients with immunohistochemical data are listed in https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf. Immunohistochemical analysis was performed to evaluate the presence and distribution of tumor-infiltrating immune elements between the best response group and the clinical benefit group (NDB; https://cdn.amegroups.cn/static/public/tlcr-2025-323-1.pdf). No significant differences were observed in TAMs or M1-like TAM between the DCB group and NDB group (CD68: P=0.06; CD86: P=0.83). However, M2-like TAM levels differed significantly between the two groups (P=0.045) (Figure 6A).

Figure 6 Physiological significance of the radiomics signatures. (A) Performance of the tumor infiltrating immune elements in assessing the clinical benefit. (B,C) Multivariate linear regression analysis revealed that the RS_WPL and RS_Lymphnode was associated with M2 like TAM. ○, represent mild outliers (values exceeding 1.5× interquartile range [IQR] from the quartiles); *, denote extreme outliers (values exceeding 3× IQR). DCB, durable clinical benefit; IQR, interquartile range; NCB, no durable benefit; RS, radiomic signature; TAM, tumor-associated macrophage; WPL, whole tumor, peritumor and lymph node.

Table 3 presents the correlations between radiomics signatures and tumor-infiltrating immune elements, based on criteria described in the Materials and Methods section. Pearson correlation analysis showed that RS_Lymph and RS_WPL were negatively correlated with M2-like TAMs (ρ for RS_Lymph =−0.367, P=0.01; ρ for RS_WPL =−0.519, P<0.01) (Figure 6B,6C). In contrast, RS_WT, RS_Peri, and RS_WP showed no association with M2-like TAMs.

Multivariate linear regression analysis (adjusted R²=0.20; F=10.70; P=0.002) further indicated that RS_WPL was independently associated with M2-like TAM (β=−0.4, P=0.002), accounting for 20.0% of the variability in RS_WPL (Figure 6B).

Discussion

In this study, we built a radiomics signature based on the primary tumors, peri-tumors and lymph nodes to simultaneously predict multiple clinical outcomes in immunotherapy-treated NSCLC patients, including DCB, PFS, OS, best response and pseudoprogression. This approach helps clinicians make timely and effective personalized clinical decisions based on the complexity of immunotherapy. Additionally, we identified M2-like TAMs as a valuable biomarker for further evaluation. The diverse roles of M2-like TAMs in both typical and atypical immune therapy responses underpin the physiological significance of RS_WPL in predicting different clinical outcomes, while also highlighting the importance and necessity of incorporating lymph nodes features.

The RS_WPL we developed can predict multiple functions while maintaining diagnostic performance across various predictive functions. As a predictive factor of DCB, RS_WPL achieved a high AUC (0.84) and sensitivity (0.84), results comparable to those of Trebeschi et al., who used machines learning to predict the prognosis of lung cancer immunotherapy responses, achieving an AUC of 0.83 (23). Our conclusion shows a slightly higher AUC (0.80) compared to the study by Yang et al., which extracted peritumoral and intratumoral radiomics features (17). However, our AUC is slightly lower than the results obtained by Khorrami et al., who also extracted intratumoral and peritumoral texture features (AUC =0.88) (22). This discrepancy may be due to the fact that we only utilized pre-treatment CT scans, whereas they employed the difference between pre- and post-treatment scans. This suggests that we have the potential to identify patients who may benefit from ICIs treatment before the therapy begins.

Beyond predictive performance, the interpretability of RS_WPL was further supported by the characteristics of its selected features, which were extracted from three anatomically distinct regions: the tumor area, peritumoral margins, and lymph nodes.

The tumor features included 3D Wavelet P2 L2 C15, Center of Mass Shift, Compactness 2, and GLSZM_Large zone low grey level emphasis. These features characterize intratumoral heterogeneity, structural irregularity, and grey-level distribution. Wavelet-based features, such as 3D Wavelet P2 L2 C15, have shown high discriminative power in cancer classification tasks and were top-ranked in prior feature selection processes (31). Compactness 2 reflects tumor shape complexity and was proven useful in lung cancer diagnostic SVM models (32). Center of Mass Shift represents the spatial shift in voxel intensity distribution, potentially reflecting asymmetrical tumor growth. GLSZM_Large zone low grey level emphasis captures broad low-intensity areas within the tumor, which may correlate with hypoxia or necrosis (33).

Peritumoral features, extracted from 5mm and 10mm extended margins, included 3D Wavelet P1 L2 C13, GLSZM_Zone size non-uniformity, Statistical ENERGY, Intensity Histogram Quartile Dispersion, and Laws features. These capture texture heterogeneity, intensity dispersion, and local morphological patterns in the tumor microenvironment. The peritumoral region serves as the interface between tumor and normal tissue, where microenvironmental changes relate closely to tumor invasion, local immune response, and angiogenesis. Wavelet and GLSZM features detect subtle structural heterogeneity; intensity histogram and variance metrics quantify intensity variability, Laws features identify microstructural texture patterns, and Statistical ENERGY measures texture uniformity (34-36). Collectively, these features reveal the complex biology of the peritumoral microenvironment, supporting their prognostic relevance. The RS_WP combined tumor radiomics signature and peritumoral radiomics signature, enabling a more comprehensive characterization of intra- and peritumoral heterogeneity (37,38), thereby achieving superior predictive performance.

Lymph node features, a unique and novel component of RS_WPL, included GLSZM_Zone size non-uniformity normalized, minimum histogram gradient, and 3D Laws (L5 L5 R5, L5 R5 L5), revealing the complexity of texture structure and morphological changes in the lymph nodes. As critical sites for tumor metastasis and immune response, the inclusion of these features significantly enhanced the model’s ability to capture the overall disease status.

In summary, the integration of multi-scale, multi-region radiomic features allows RS_WPL to simultaneously capture tumor burden, microenvironmental dynamics, and immune involvement. This spatially stratified modeling approach provides greater biological interpretability and significantly improves prognostic prediction compared to tumor-localized features alone.

In this study, the incidence rates of pseudoprogression and hyperprogression were 5.7% and 18.4%, respectively, with no significant differences in their clinical characteristics and onset times, consistent with the findings of Nishino et al. (39)]. He et al. (40) utilized intratumoral and peritumoral features to distinguish pseudoprogression from hyperprogression, achieving an AUC of 0.834. Vaidya et al. (41) employed machine learning to distinguish between HP and other radiographical response patterns with an AUC of 0.85±0.06. Our study demonstrated that RS_WPL could differentiate pseudoprogression patients from hyperprogression patients, with an AUC of 0.80. Although our results were slightly lower than the previous two studies, RS_WPL is based on the simultaneous realization of multiple predictive functions, including DCB, whereas the aforementioned studies focused on models specifically for pseudoprogression and hyperprogression patients, resulting in more singular predictive outcomes.

Lymph nodes play a significant role in the prognosis of lung cancer, with more lymph node metastasis sites leading to a worse prognosis (42). Our study demonstrates that the good prediction of multiple outcomes is mainly due to the addition of lymph node feature extraction. This is primarily reflected in two aspects. On one hand, in this study, the predictive ability of the individual omics features of the primary tumor and lymph nodes for sustained benefit from immunotherapy is the same (AUC =0.71), which is higher than that of the peritumoral features (RSPeri5mm, AUC =0.69; RSPeri10mm, AUC =0.64). The combined RS_WPL has better prognostic performance (AUC =0.84). On the other hand, in this study, RS_Lymph and RS_WPL are associated with M2-like TAMs in predicting disease control rate (DCR) and progression-free rate (PFR), while no significant correlation was found between RS_WT and RS_Peri and M2-like TAM. Our study further reveals that RS_WPL is independently associated with M2-like TAM (coefficient =−0.4; P=0.002), contributing 20.0% to the variability of RS_WPL. This highlights the importance and necessity of including lymph nodes. Basic research (25,43,44) has found that M2-like TAM can induce angiogenesis and lymphangiogenesis around tumors in mice, and subsequent clinical studies have also confirmed that M2-like TAM can promote tumor progression, invasion, and lymph node metastasis in humans. Additionally, M2-like TAM is enriched in high-progression patients, and TAM reprogramming can induce high progression (3). These studies demonstrate that TAM, especially M2-like polarization, plays a role in various typical and atypical responses to immunotherapy and also reveals the physiological significance of RS_WPL in predicting diverse clinical outcomes. Finally, TAMs in malignant tumors are new targets for tumor immunotherapy, and new immunotherapies targeting macrophages have been applied in clinical settings (45). The implementation of multi-modal imaging omics to predict M2-like TAMs can stratify patients undergoing TAM-modulating therapies and monitor their responses to these treatments.

We acknowledged that our study has several limitations. First, the study highly relies on RECIST as prediction targets. We would acknowledge RECIST limitations in this sense. Here, we would also highlight the inherent limitations of RECIST, particularly the notably high interobserver variability, which is attributed to factors such as measurement discrepancies and target lesion selection. Second, although we performed an external evaluation of prognostic prediction for different immunetherapy responses, the study cohort remains relatively small. Moving forward, we plan to conduct a prospective study with a larger patient sample to validate our current findings. Additionally, we recognize that the subsets involving pseudoprogression and hyperprogression were small, which constrains the statistical power of our analysis. This is because these phenomena are relatively rare—consistent with previous reports (18,46)—with proportions of less than 8% for pseudoprogression and 10% for hyperprogression, respectively, further limiting the sample size available for analysis. Next, we will continue to collect more cases to strengthen validation. Furthermore, the study employed different TAMs phenotypes as the sole explanatory variable in the model, which may limit the biological relevance of the model since TAM is not the only predictor of patient response to immunotherapy.

Conclusions

We constructed an effective and stable radiomics signature by integrating tumor, peritumoral region, and lymph node features derived from baseline CT images, which can serve as a predictive biomarker for different immunotherapy responses and prognosis. The relationship between this radiomics signature and M2-like TAM not only explains the biological significance of the radiomics model but also assists clinicians in making informed clinical decisions.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-323/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-323/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-323/prf

Funding: This study was supported by the National Natural Science Foundation of China (No. 62176013 to W.M.) and the Shandong Provincial Natural Science Foundation (No. ZR2024QF010 to W.M.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-323/coif). W.M. reports the funding from the National Natural Science Foundation of China (No. 62176013) and the Shandong Provincial Natural Science Foundation (No. ZR2024QF010). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments, and was approved by the Institutional Review Board of Shandong Cancer Hospital and Institute, Shandong Provincial Hospital Affiliated with Shandong First Medical University (No. SDTHEC2020001015). Owing to the retrospective design, the requirement for informed consent was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Rizvi NA, Mazières J, Planchard D, et al. Activity and safety of nivolumab, an anti-PD-1 immune checkpoint inhibitor, for patients with advanced, refractory squamous non-small-cell lung cancer (CheckMate 063): a phase 2, single-arm trial. Lancet Oncol 2015;16:257-65. [Crossref] [PubMed]
2. Garon EB, Rizvi NA, Hui R, et al. Pembrolizumab for the treatment of non-small-cell lung cancer. N Engl J Med 2015;372:2018-28. [Crossref] [PubMed]
3. Lo Russo G, Moro M, Sommariva M, et al. Antibody-Fc/FcR Interaction on Macrophages as a Mechanism for Hyperprogressive Disease in Non-small Cell Lung Cancer Subsequent to PD-1/PD-L1 Blockade. Clin Cancer Res 2019;25:989-99. [Crossref] [PubMed]
4. Saâda-Bouzid E, Defaucheux C, Karabajakian A, et al. Hyperprogression during anti-PD-1/PD-L1 therapy in patients with recurrent and/or metastatic head and neck squamous cell carcinoma. Ann Oncol 2017;28:1605-11. [Crossref] [PubMed]
5. Champiat S, Dercle L, Ammari S, et al. Hyperprogressive Disease Is a New Pattern of Progression in Cancer Patients Treated by Anti-PD-1/PD-L1. Clin Cancer Res 2017;23:1920-8. [Crossref] [PubMed]
6. Kanjanapan Y, Day D, Wang L, et al. Hyperprogressive disease in early-phase immunotherapy trials: Clinical predictors and association with immune-related toxicities. Cancer 2019;125:1341-9. [Crossref] [PubMed]
7. Ferrara R, Mezquita L, Texier M, et al. Hyperprogressive Disease in Patients With Advanced Non-Small Cell Lung Cancer Treated With PD-1/PD-L1 Inhibitors or With Single-Agent Chemotherapy. JAMA Oncol 2018;4:1543-52. [Crossref] [PubMed]
8. Tunali I, Gray JE, Qi J, et al. Novel clinical and radiomic predictors of rapid disease progression phenotypes among lung cancer patients treated with immunotherapy: An early report. Lung Cancer 2019;129:75-9. [Crossref] [PubMed]
9. Sasaki A, Nakamura Y, Mishima S, et al. Predictive factors for hyperprogressive disease during nivolumab as anti-PD1 treatment in patients with advanced gastric cancer. Gastric Cancer 2019;22:793-802. [Crossref] [PubMed]
10. Ott PA, Hodi FS, Robert C. CTLA-4 and PD-1/PD-L1 blockade: new immunotherapeutic modalities with durable clinical benefit in melanoma patients. Clin Cancer Res 2013;19:5300-9. [Crossref] [PubMed]
11. Gandhi L, Rodríguez-Abreu D, Gadgeel S, et al. Pembrolizumab plus Chemotherapy in Metastatic Non-Small-Cell Lung Cancer. N Engl J Med 2018;378:2078-92. [Crossref] [PubMed]
12. Brahmer J, Reckamp KL, Baas P, et al. Nivolumab versus Docetaxel in Advanced Squamous-Cell Non-Small-Cell Lung Cancer. N Engl J Med 2015;373:123-35. [Crossref] [PubMed]
13. Herbst RS, Soria JC, Kowanetz M, et al. Predictive correlates of response to the anti-PD-L1 antibody MPDL3280A in cancer patients. Nature 2014;515:563-7. [Crossref] [PubMed]
14. Dammeijer F, van Gulijk M, Mulder EE, et al. The PD-1/PD-L1-Checkpoint Restrains T cell Immunity in Tumor-Draining Lymph Nodes. Cancer Cell 2020;38:685-700.e8. [Crossref] [PubMed]
15. Rahim MK, Okholm TLH, Jones KB, et al. Dynamic CD8(+) T cell responses to cancer immunotherapy in human regional lymph nodes are disrupted in metastatic lymph nodes. Cell 2023;186:1127-1143.e18. [Crossref] [PubMed]
16. He B, Dong D, She Y, et al. Predicting response to immunotherapy in advanced non-small-cell lung cancer using tumor mutational burden radiomic biomarker. J Immunother Cancer 2020;8:e000550. [Crossref] [PubMed]
17. Yang B, Zhou L, Zhong J, et al. Combination of computed tomography imaging-based radiomics and clinicopathological characteristics for predicting the clinical benefits of immune checkpoint inhibitors in lung cancer. Respir Res 2021;22:189. [Crossref] [PubMed]
18. Queirolo P, Spagnolo F. Atypical responses in patients with advanced melanoma, lung cancer, renal-cell carcinoma and other solid tumors treated with anti-PD-1 drugs: A systematic review. Cancer Treat Rev 2017;59:71-8. [Crossref] [PubMed]
19. Shen L, Fu H, Tao G, et al. Pre-Immunotherapy Contrast-Enhanced CT Texture-Based Classification: A Useful Approach to Non-Small Cell Lung Cancer Immunotherapy Efficacy Prediction. Front Oncol 2021;11:591106. [Crossref] [PubMed]
20. Polverari G, Ceci F, Bertaglia V, et al. (18)F-FDG Pet Parameters and Radiomics Features Analysis in Advanced Nsclc Treated with Immunotherapy as Predictors of Therapy Response and Survival. Cancers (Basel) 2020;12:1163. [Crossref] [PubMed]
21. Mu W, Tunali I, Gray JE, et al. Radiomics of (18)F-FDG PET/CT images predicts clinical benefit of advanced NSCLC patients to checkpoint blockade immunotherapy. Eur J Nucl Med Mol Imaging 2020;47:1168-82. [Crossref] [PubMed]
22. Khorrami M, Prasanna P, Gupta A, et al. Changes in CT Radiomic Features Associated with Lymphocyte Distribution Predict Overall Survival and Response to Immunotherapy in Non-Small Cell Lung Cancer. Cancer Immunol Res 2020;8:108-19. [Crossref] [PubMed]
23. Trebeschi S, Drago SG, Birkbak NJ, et al. Predicting response to cancer immunotherapy using noninvasive radiomic biomarkers. Ann Oncol 2019;30:998-1004. [Crossref] [PubMed]
24. Mantovani A. La mala educación of tumor-associated macrophages: Diverse pathways and new players. Cancer Cell 2010;17:111-2. [Crossref] [PubMed]
25. Zhang B, Yao G, Zhang Y, et al. M2-polarized tumor-associated macrophages are associated with poor prognoses resulting from accelerated lymphangiogenesis in lung adenocarcinoma. Clinics (Sao Paulo) 2011;66:1879-86. [Crossref] [PubMed]
26. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45:228-47. [Crossref] [PubMed]
27. Rizvi NA, Hellmann MD, Snyder A, et al. Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer. Science 2015;348:124-8. [Crossref] [PubMed]
28. Campesato LF, Barroso-Sousa R, Jimenez L, et al. Comprehensive cancer-gene panels can be used to estimate mutational load and predict clinical benefit to PD-1 blockade in clinical practice. Oncotarget 2015;6:34221-7. [Crossref] [PubMed]
29. Kato S, Goodman A, Walavalkar V, et al. Hyperprogressors after Immunotherapy: Analysis of Genomic Alterations Associated with Accelerated Growth Rate. Clin Cancer Res 2017;23:4242-50. [Crossref] [PubMed]
30. Di Giacomo AM, Danielli R, Guidoboni M, et al. Therapeutic efficacy of ipilimumab, an anti-CTLA-4 monoclonal antibody, in patients with metastatic melanoma unresponsive to prior systemic treatments: clinical and immunological evidence from three patient cases. Cancer Immunol Immunother 2009;58:1297-306. [Crossref] [PubMed]
31. Krewer H, Geiger B, Hall LO, et al. Effect of texture features in computer aided diagnosis of pulmonary nodules in low-dose computed tomography. 2013 IEEE International Conference on Systems, Man, and Cybernetics. Manchester: IEEE; 2013:3887-91.
32. Nawreen N, Hany U, Islam T. Lung cancer detection and classification using CT scan image processing. 2021 International Conference on Automation, Control and Mechatronics for Industry 4.0 (ACMI). Rajshahi: IEEE; 2021:1-6.
33. Homayounieh F, Ebrahimian S, Babaei R, et al. CT Radiomics, Radiologists, and Clinical Information in Predicting Outcome of Patients with COVID-19 Pneumonia. Radiol Cardiothorac Imaging 2020;2:e200322. [Crossref] [PubMed]
34. Khorrami M, Khunger M, Zagouras A, et al. Combination of Peri- and Intratumoral Radiomic Features on Baseline CT Scans Predicts Response to Chemotherapy in Lung Adenocarcinoma. Radiol Artif Intell 2019;1:e180012. [Crossref] [PubMed]
35. Beig N, Khorrami M, Alilou M, et al. Perinodular and Intranodular Radiomic Features on Lung CT Images Distinguish Adenocarcinomas from Granulomas. Radiology 2019;290:783-92. [Crossref] [PubMed]
36. Tunali I, Hall LO, Napel S, et al. Stability and reproducibility of computed tomography radiomic features extracted from peritumoral regions of lung cancer lesions. Med Phys 2019;46:5075-85. [Crossref] [PubMed]
37. Hu Y, Xie C, Yang H, et al. Assessment of Intratumoral and Peritumoral Computed Tomography Radiomics for Predicting Pathological Complete Response to Neoadjuvant Chemoradiation in Patients With Esophageal Squamous Cell Carcinoma. JAMA Netw Open 2020;3:e2015927. [Crossref] [PubMed]
38. Bao X, Peng Q, Bian D, et al. Short-term intra- and peri-tumoral spatiotemporal CT radiomics for predicting major pathological response to neoadjuvant chemoimmunotherapy in non-small cell lung cancer. Eur Radiol 2025;35:6052-64. [Crossref] [PubMed]
39. Nishino M, Ramaiya NH, Chambers ES, et al. Immune-related response assessment during PD-1 inhibitor therapy in advanced non-small-cell lung cancer patients. J Immunother Cancer 2016;4:84. [Crossref] [PubMed]
40. He S, Feng Y, Lin Q, et al. CT-Based Peritumoral and Intratumoral Radiomics as Pretreatment Predictors of Atypical Responses to Immune Checkpoint Inhibitor Across Tumor Types: A Preliminary Multicenter Study. Front Oncol 2021;11:729371. [Crossref] [PubMed]
41. Vaidya P, Bera K, Patil PD, et al. Novel, non-invasive imaging approach to identify patients with advanced non-small cell lung cancer at risk of hyperprogressive disease with immune checkpoint blockade. J Immunother Cancer 2020;8:e001343. [Crossref] [PubMed]
42. Maniwa T, Ohmura A, Hiroshima T, et al. Number of metastatic lymph nodes and zones as prognostic factors in non-small-cell lung cancer. Interact Cardiovasc Thorac Surg 2020;31:305-14. [Crossref] [PubMed]
43. Zhang B, Wang J, Gao J, et al. Alternatively activated RAW264.7 macrophages enhance tumor lymphangiogenesis in mouse lung adenocarcinoma. J Cell Biochem 2009;107:134-43. [Crossref] [PubMed]
44. Dirkx AE, Oude Egbrink MG, et al. Monocyte/macrophage infiltration in tumors: modulators of angiogenesis. J Leukoc Biol 2006;80:1183-96. [Crossref] [PubMed]
45. Genard G, Lucas S, Michiels C. Reprogramming of Tumor-Associated Macrophages with Anticancer Therapies: Radiotherapy versus Chemo- and Immunotherapies. Front Immunol 2017;8:828. [Crossref] [PubMed]
46. Hodi FS, Hwu WJ, Kefford R, et al. Evaluation of Immune-Related Response Criteria and RECIST v1.1 in Patients With Advanced Melanoma Treated With Pembrolizumab. J Clin Oncol 2016;34:1510-7. [Crossref] [PubMed]