Lung cancer, which originates primarily from epithelial cells in the bronchial mucosa or alveoli, is the leading cause of cancer-related morbidity and is often accompanied by an unfavorable prognosis. Among its subtypes, non-small cell lung cancer (NSCLC) is predominant, comprising lung adenocarcinoma and squamous cell carcinoma (1). Although surgical resection remains the cornerstone of NSCLC management, its applicability diminishes in the advanced stages owing to the associated poor prognosis. For patients with advanced stages of NSCLC when surgery is not viable, therapeutic efforts focus on prolonging survival and increasing the quality of life, aiming for sustained survival despite the presence of tumor (2). Current prognostic markers for lung cancer include clinicopathological characteristics, laboratory diagnostic markers such as carcinoembryonic antigen and carbohydrate antigen 125 (3), and CT imaging modalities. However, these markers do not meet the evolving demands of precision medicine, underscoring the need to discover novel prognostic biomarkers that can guide personalized precision interventions.

Natural killer (NK) cells, pivotal for the body’s innate immune arsenal, possess the capability to eliminate target cells without antigen-specific stimulation and play instrumental roles in immune clearance and surveillance. Their role in targeting and eradicating tumor cells has been well established. For example, in hepatocellular carcinoma, curtailing antitumor sphingomyelin synthesis in peripheral NK cells can delineate alterations in the membrane structure of intratumoral NK cells, paving the way for innovative therapeutic strategies (4). Furthermore, in breast cancer, the interaction between HLA-G and the NK cell receptor KIR2DL4 amplifies the susceptibility of HER2-positive breast cancer cells to trastuzumab, elucidating the mechanisms underlying the resistance to trastuzumab (5). Studies have revealed that inhibition of IL6 can potentiate NK cell-mediated cytotoxicity in osimertinib-resistant EGFR-mutant NSCLC cells (6). Furthermore, another study demonstrated that the upregulation of SERPINB4 in NSCLC dampens NK cell-mediated cytotoxicity, suggesting potential therapeutic avenues for modulating immune responses against NSCLC (7). These findings emphasize the pivotal role of NK cells in orchestrating antitumor responses in NSCLC and other malignancies and highlight novel avenues for NK cell-centric therapeutic strategies.

Evaluation of NK cell infiltration is based predominantly on flow cytometry using fresh tissue specimens. Although effective, this technique has challenges, such as real-time detection constraints and elevated costs. Alternative methodologies include the use of public databases, such as ImmuCellAI and Cibersortx. However, these are contingent upon the availability of the corresponding genomic data and are susceptible to variables, such as specimen collection methods and antibody specificity. Imaging examinations continue to be a linchpin in clinical diagnostics because they are impervious to real-time variables, cost-effective, and widely accessible. This paved the way for radiomics, a discipline that uses medical imaging for research.

Recently, the integration of radiomics into medical research has increased. Radiomics creates models by extracting image features imperceptible to the human eye. When combined with clinical and pathological datasets, these models can help design diagnostic and prognostic frameworks. The spectrum of application of radiomics includes tumor diagnosis, post-therapeutic evaluations, and prognostic evaluations (8). Recent studies have demonstrated the versatility and effectiveness of CT-based radiomics in cancer diagnosis and treatment. One study (9) found that changes in CT numbers and tumor volume from cone-beam computed tomography can predict early response in NSCLC treatment, highlighting significant differences between responding and non-responding patients. Further, another study (10) showed that combining CT-based radiomics features with clinical factors can effectively predict PD-L1 and CD8+TILs expression levels in esophageal squamous cell carcinoma, thereby enhancing the accuracy of predictions. Similarly, a study (11) revealed that CT-based radiomics, when combined with clinical and morphological factors, can accurately predict PD-L1 expression levels and tumor mutation burden in advanced-stage NSCLC, which is crucial for guiding immunotherapy. Another research effort (12) demonstrated the potential of CT-based radiomics, in conjunction with clinical and radiological parameters, to effectively differentiate between immune checkpoint inhibitor-related pneumonitis and radiation pneumonitis in patients with advanced-stage NSCLC. These studies collectively underscore the growing importance of radiomics in oncological imaging and personalized medicine. In NSCLC, radiomic models enriched with multiregional attributes have been crucial in stratifying survival risks for patients diagnosed with clinical and pathological stage IA disease (13). A seminal study by Tong et al. (14) innovatively used machine learning algorithms to predict the expression levels of CD8 in tumor-infiltrating immune cells in NSCLC, leveraging PET/CT radiomics and clinical attributes, offering insights into the nuances of the tumor microenvironment. However, the radiomic domain has not yet been explored for the evaluation of NK cell infiltration in NSCLC.

Therefore, this study paved the way for the application of chest CT images in the construction of a radiomics model. The model is based on machine learning algorithms and aims to predict the abundance of NK cell infiltration in NSCLC tissues without invasive methods. By amalgamating bioinformatic analyses, our endeavor seeks to explore the molecular intricacies governing NK cell expression in NSCLC and their interaction with the immune microenvironment, which can provide novel perspectives for the diagnosis, therapeutic interventions, and prognostic evaluation of patients with NSCLC.

Once metastasis occurs in NSCLC, with a 5-year relative survival rate of only 5%, the prognosis of NSCLC is poor (17, 18). Therefore, new prognostic predictors for the precise treatment of NSCLC are urgently needed. In this study, we analyzed the association between the histological characteristics of NSCLC and NK cells in a tumor microenvironment. This study aimed to predict NK cell infiltration in NSCLC noninvasively to improve prognosis prediction for patients. The predictive probability of the LR radiomics model for the abundance of NK cells agreed with the true value, and the model predictive value of RS positively correlated with the infiltration of NK cells. The nomogram model, which was established by combining clinical features (T, N, and M stages) with the RS of the radiomics model, showed a good predictive ability for patient prognosis.

In this study, to exclude the impact of clinical factors on the expression of NK cells and the prognosis of lung cancer, we employed a multivariate Cox regression analysis. After adjusting for the influence of 10 clinical demographic confounders, we found that NK cell expression remains a protective factor in the prognosis of lung cancer (OR: 0.717, 95% CI: 0.558-0.922, p=0.01). Additionally, through subgroup analysis and interaction tests, it was observed that there is no interaction between NK cell expression and the clinical demographic characteristics above in terms of the prognosis of lung cancer.

Several challenges remain in the contemporary field of NSCLC management. Despite the increased adoption of molecular targeted therapies and immunotherapies, a subset of patients, especially those with advanced disease stages, exhibit limited therapeutic responses (19). Despite their potential, the emergence of innovative diagnostic modalities faces barriers, such as high costs and intricate technical demands that limit their widespread adoption (20). Existing therapeutic regimens have yet to mitigate the associated adverse effects and complications, compromising patients’ quality of life.

NK cells play a role in many physiologic and pathologic processes in NSCLC. For example, NK cells in MHC class I-deficient lung adenocarcinoma displayed impaired cytotoxic activity towards tumor cells, associated with alterations in the expression of natural cytotoxicity receptors and ligands required for target cell recognition (21). NK cells produce cytokines, such as IFN-γ and TNF-α, which have anti-tumor effects. These cytokines can enhance the immune response by recruiting and activating other immune cells, including T cells, to the tumor site (22–24). NK cells interact with dendritic cells, macrophages, and T cells, modulating the immune response. And they can also shape the tumor microenvironment, influencing the efficacy of other immunotherapeutic strategies (25–27).

Recent advances in NSCLC research have highlighted the role of NK cells. These preliminary findings underscore the therapeutic potential of NK cells for NSCLC immunotherapy (28). To capitalize on the recognition of innate tumor antigens and cytotoxic capabilities of NK cells, ongoing research has aimed to improve their antitumor efficacy (29). Evidence suggests that specific methodologies can enhance NK cell activity and enhance tumor cell recognition (30). The intricate interplay between NK and lung cancer cells has emerged as a key research focus. Studies have shown that NK cells exert their cytotoxic effects by recognizing and engaging with specific receptors (31). The ability of the lung cancer microenvironment to modulate NK cell functionality and potentially attenuate its antitumor responses is a subject of intense scrutiny. Certain microenvironmental factors may impede NK cell functionality, diminishing their antitumor effects (32). Determining inhibitory mechanisms and strategizing countermeasures are critical research avenues.

In immunotherapeutic strategies, NK cells also play an important role. Many tumors, including NSCLC, exploit inhibitory pathways to escape NK cell-mediated killing (33, 34). New therapies aim to block these inhibitory signals (like KIRs, NKG2A) to enhance NK cell activity (35, 36). Strategies that boost activating signals on NK cells can enhance their ability to target NSCLC cells (37). This includes the use of antibodies that engage activating receptors like NKG2D or the use of cytokines like IL-15 to stimulate NK cell activity. NK cell activity and related markers are being explored as potential biomarkers to predict response to immunotherapy in NSCLC (38, 39).

In our study, we identified a positive association between NK cell infiltration and the expression of the ICOS (Inducible T-cell COStimulator) and BTLA (B and T Lymphocyte Attenuator) genes. A pronounced correlation was also observed between M0 and M2 macrophage infiltration. ICOS is a co-stimulatory molecule found on T cells. It plays a crucial role in T cell activation and survival. ICOS is part of the CD28 superfamily and interacts with its ligand, ICOSL, found on antigen-presenting cells (40–42). BTLA is an immune checkpoint molecule expressed on T cells, B cells, and to a lesser extent on NK cells and macrophages (43). It negatively regulates immune responses by binding to its ligand, HVEM (Herpesvirus Entry Mediator) (44–46).

ICOS and BTLA are associated with a variety of immune cells such as NK cells, M0 and M2 macrophages (47, 48). Firstly, NK cells typically do not express ICOS under normal physiological conditions. However, certain conditions, like chronic viral infections or certain cancer microenvironments, might induce ICOS expression on NK cells, potentially altering their function (49). NK cells can express BTLA. The engagement of BTLA on NK cells generally leads to an inhibitory signal, potentially reducing their cytotoxic function (50). The expression and functional impact of BTLA on NK cells can vary depending on the disease context and the microenvironment (50).

M0 Macrophages are considered to be in a resting state and can polarize into either M1 (pro-inflammatory) or M2 (anti-inflammatory) phenotypes based on the microenvironment (51, 52). M2 macrophages are associated with tissue repair, wound healing, and immune regulation (51). The role of ICOS and BTLA in M2 macrophages is less clear. However, given that BTLA is an inhibitory molecule, its expression in the macrophage context could be associated with the suppressive functions of M2 macrophages (53).

Gene enrichment analysis revealed that genes associated with NK cell immune infiltration were predominantly aligned with the p53 and Wnt signaling pathways. Thus, these insights will serve as a foundation for subsequent genomic and transcriptional regulation studies centered on NK cells.

By examining the gene expression profiles and signaling cascades of NK cells, we aim to elucidate the molecular dynamics of the participation of NK cells in NSCLC, informing the development of innovative therapeutic paradigms. In recent years, radiomics technology has become an important direction in NSCLC research. Radiomics analyzes medical imaging data using high-throughput methods that can reveal the microscopic and macroscopic features of tumors, thus contributing to a better understanding and treatment of NSCLC (54). First, histological imaging techniques have been effective in aiding the early diagnosis and staging of NSCLC (55). By analyzing specific imaging markers in CT or MRI, researchers can more accurately determine the size, location, and aggressiveness of a tumor, which can help develop more targeted treatment plans (56). Furthermore, histological imaging can predict the prognosis of a patient by analyzing the morphology and textural characteristics of the tumor. However, radiomics technology has certain shortcomings that limit its application in clinical practice (57). With the continuous development of multiomics technology, imaging histology technology is expected to further improve its application value in NSCLC research and treatment when combined with other bioinformatics methods. For example, by combining genomic and proteomic data, imaging genomics can provide a more comprehensive and in-depth tumor analysis, which can help to discover new therapeutic targets and strategies (56). Currently, there are no reports on radiomics and NK cells. Here, the radiomics model based on the features original_gldm_SmallDependenceHighGrayLevelEmphasis and original_gldm_LowGrayLevelEmphasis had a positive effect on NK cell infiltration in NSCLC and patient prognosis predictive effect, and the predicted probability of high and low NK cell infiltration abundance agreed with the true value. To the best of our knowledge, this is the first radiomics study to predict NK cell infiltration and patient prognosis in NSCLC. The present study fills a gap in imaging histology in NSCLC research and could facilitate the prediction of NK cell infiltration levels in NSCLC. The study also explored the genes, immune infiltration, and pathways associated with the abundance of NK cell infiltration by combining bioinformatics technology, which provides a new way to understand the molecular mechanism behind the imaging histology model in a more detailed and comprehensive manner.

Although the image histology model constructed in this study worked well, it had some shortcomings. First, data for this study were obtained from public databases with non-uniform image parameters. Second, the sample size was limited and needs to be further validated using a large-sample multicenter study. Finally, this study used a manual explanation of VOIs, which is also affected by subjective factors.

The radiomics model constructed in this study using machine learning algorithms based on enhanced CT images has good stability and diagnostic efficacy and can accurately predict the infiltration of NK cells in NSCLC. Furthermore, the comprehensive model established by combining relevant clinical characteristics can predict the prognosis of patients with NSCLC. Moreover, we explored the potential gene expression and regulation of the pathway underlying NK cell infiltration in NSCLC by combining bioinformatic technology, which provides a new strategy to assist in clinical decision-making and guide individualized precision diagnosis and treatment of NSCLC. In the near future, the combination of multicenter data and well-established mechanistic studies could present novel opportunities for the treatment of patients with NSCLC.

Publicly available datasets were analyzed in this study. This data can be found here: https://portal.gdc.cancer.gov/.

XM: Conceptualization, Data curation, Validation, Writing – original draft. HX: Methodology, Writing – original draft. YL: Data curation, Writing – original draft. ML: Writing – review & editing. WS: Writing – review & editing. BZ: Writing – review & editing. JS: Writing – review & editing. MD: Writing – review & editing. YG: Investigation, Funding acquisition, Writing – review & editing.