Lymph node metastasis in early gastric cancer: A review of risk stratification and surgical decision-making

Abstract

Early gastric cancer usually has a favorable prognosis, and endoscopic submucosal dissection allows many low-risk patients to avoid invasive surgery. This changes, however, when lymph node metastasis occurs; in these patients, gastrectomy with lymph node dissection is often required, which results in greater surgical trauma and a higher risk of postoperative functional impairment. However, preoperative assessment of lymph node metastasis in early gastric cancer remains challenging. Current diagnostic approaches provide useful information, yet no single method can reliably exclude small or occult nodal metastases. Therefore, this review focuses on the risk assessment of lymph node metastasis as a central issue in early gastric cancer. It discusses the biological basis of nodal spread, the limits of current diagnostic strategies, and the influence of these uncertainties on treatment selection. An integrated use of clinicopathological, imaging, and molecular information may support more individualized management of early gastric cancer.

Key Words: Early gastric cancer; Lymph node metastasis; Risk assessment; Imaging diagnosis; Artificial intelligence

Core Tip: Lymph node metastasis remains the central factor guiding treatment selection in early gastric cancer, but its prediction before treatment is still unreliable. This review summarizes the biological basis of lymph node metastasis, including epithelial-mesenchymal transition, lymphangiogenesis, and immune regulation, and evaluates current diagnostic approaches such as imaging, sentinel lymph node mapping, biomarkers, and artificial intelligence. The main message is that no single method can exclude lymph node metastasis with sufficient confidence. Integrated risk stratification using clinical, imaging, and molecular information is needed to support individualized treatment decisions in early gastric cancer.

INTRODUCTION

Gastric cancer accounts for a large proportion of the global cancer burden, with a global estimate of nearly one million new cases and 650000 deaths annually. This burden is unevenly distributed, with East Asia having the highest incidence[1]. Despite advances in diagnosis and treatment, prognosis remains poor once the disease has reached an advanced stage[2]. Therefore, early detection is essential as it allows for the identification of the lesion before lymph node metastasis (LNM) occurs.

Early gastric cancer (EGC) is defined based on the depth of invasion, where the lesion is restricted to the mucosa or submucosa of the stomach wall, regardless of LNM. However, this definition creates practical problems. A tumor may still be classified as “early” by depth, but its prognosis can differ greatly once LNM is present. In node-negative EGC, endoscopic submucosal dissection (ESD) is usually preferred, with favorable clinical outcomes and an overall five-year survival rate of > 90%. Once LNM is detected, the prognosis may approach that of advanced gastric cancer, and radical gastrectomy with lymph node dissection is usually required. Compared with ESD, this operation carries more postoperative complications and has a greater impact on quality of life[3].

LNM in EGC has a complex biological basis involving epithelial-mesenchymal transition (EMT), tumor-induced lymphangiogenesis, and immune-cell interactions in the tumor microenvironment (TME). This complicates preoperative identification. LNM in EGC often presents as a micrometastatic disease; therefore, conventional examinations cannot accurately exclude LNM. Artificial intelligence (AI) has emerged as a tool for integrating diagnostic information; however, current evidence remains limited and it is not yet fully clinically applied.

Therefore, this review examines LNM risk assessment in EGC by linking the biological basis of nodal spread with current diagnostic limitations and treatment decision-making. It also considers the value of integrating clinical, imaging, and molecular information to support individualized treatment of EGC. Figure 1 illustrates the relationship between tumor biology, diagnostic evaluation, and treatment decision-making in EGC.

Figure 1 Overview of lymph node metastasis in early gastric cancer. Lymph node metastasis is driven by the interaction of epithelial-mesenchymal transition, tumor-induced lymphangiogenesis, and immune regulation within the tumor microenvironment. Preoperative evaluation integrates imaging, sentinel lymph node assessment, biomarkers, and artificial intelligence models. This integrated assessment supports risk stratification and guides treatment selection, including endoscopic submucosal dissection, surgery, and laparoscopic and endoscopic cooperative surgery. LNM: Lymph node metastasis; EGC: Early gastric cancer; EMT: Epithelial-mesenchymal transition; TGF: Transforming growth factor; PI3K/Akt: Phosphatidylinositol 3’-kinase/protein kinase B; LEC: Lymphatic endothelial cells; VEGFR: Vascular endothelial growth factor receptor; IL: Interleukin; FOXP3: Forkhead box P3; Treg: Regulatory T cells; CT: Computed tomography; EUS: Endoscopic ultrasound; PET-CT: Positron emission tomography-computed tomography; miRNA: MicroRNA; AI: Artificial intelligence; ESD: Endoscopic submucosal dissection; LECS: Laparoscopic and endoscopic cooperative surgery.

BIOLOGICAL BASIS OF LYMPHATIC SPREAD IN EGC

LNM in EGC involves multiple interacting biological processes. These processes include EMT, tumor-induced lymphangiogenesis, and immune interactions within the TME. This section focuses on these components.

EMT and local invasion

The development of LNM in EGC is largely attributable to the migratory and invasive ability of tumor cells. EMT promotes these malignant behaviors by reducing epithelial features and increasing mesenchymal traits, thereby enhancing their ability to migrate and penetrate tissues. Under physiological conditions, EMT has important roles in embryogenesis, tissue repair, regenerative processes and stem cell function. Under pathological conditions, the dysregulated activation of EMT contributes to cancer initiation, progression, and metastatic dissemination in EGC[4,5].

EMT is not a binary process but a continuous spectrum with intermediate or partial EMT states[6]. The intermediate EMT state, in which cells exhibit both epithelial and mesenchymal traits, is characterized by high cellular plasticity and high metastatic potential[7]. This intermediate phenotype is associated with early metastatic dissemination and the development of LNM.

There is a decrease in the expression of epithelial markers, such as epithelial cadherin (E-cadherin) and claudin-7, and an upregulation of mesenchymal markers, such as neural cadherin and vimentin during the process of EMT. These molecular changes are directed by specific transcription factors, such as Snail, Slug, zinc finger E-box binding homeobox 1/2, and Twist, which simultaneously induce mesenchymal gene expression and inhibit epithelial characteristics[5,8]. EMT is also regulated by various intracellular signaling pathways, including transforming growth factor-beta (TGF-β), wingless-related integration site (Wnt)/β-catenin, and phosphatidylinositol 3’-kinase/protein kinase B (PI3K/Akt). These signaling pathways do not act in isolation but form a tightly linked system that integrates various regulatory factors[9]. Zhang et al[10] showed that forkhead box (FOX) Q1 regulates LNM in gastric cancer through the regulation of Snail. The chaperonin-containing T-complex protein 1 subunit has also been reported to induce EMT through the Wnt/β-catenin pathway and enhance LNM in EGC[11]. Through these molecular changes, EMT enhances the migratory and invasive abilities of gastric cancer cells, allowing the cells to breach the basement membrane, infiltrate surrounding connective tissue, and ultimately enter lymphatic routes[12]. Figure 2 illustrates the EMT-related processes in EGC.

Figure 2 Schematic representation of epithelial-mesenchymal transition-driven lymph node metastasis in early gastric cancer. Epithelial-mesenchymal transition is characterized by loss of epithelial markers and gain of mesenchymal markers, accompanied by activation of transcription factors such as Snail, Slug, zinc finger E-box binding homeobox 1/2, and Twist. These changes enhance tumor cell motility and invasiveness, facilitating tumor cell dissemination and lymphatic spread. ZEB1/2: Zinc finger E-box binding homeobox 1/2; PI3K/Akt: Phosphatidylinositol 3’-kinase/protein kinase B; TGF: Transforming growth factor; EMT: Epithelial-mesenchymal transition.

Although EMT promotes tumor cell invasion, it does not explain the differences in LNM patterns among patients with similar tumors. These differences are more likely shaped by the interaction between EMT, tumor-induced lymphangiogenesis, and the microenvironment through shared signaling pathways, such as TGF-β and PI3K/Akt.

Tumor-induced lymphangiogenesis and lymphatic access

LNM in gastric cancer is closely associated with the gastric lymphatic system. The gastric lymphatic network begins in the superficial layer of the muscularis mucosae, including the intramural lymphatic system and submucosal vascular plexus, and extends through all layers of the gastric wall[13]. The lymphatic vessels of the mucosa form a common drainage pathway to the submucosa, from which they drain either directly or via communicating branches. The lymphatic vessels within the muscularis mucosae are irregularly distributed and run in a complex manner between the muscle bundles. Numerous submucosal lymphatic vessels penetrate the muscle bundles and, once they reach the serosa, form a dense lymphatic network and complex extramural lymphatic drainage pathway[14-16]. Therefore, cancer cells can enter the lymphatic system early. The distribution of lymph nodes around the stomach and lymphatic vessels in the gastric wall is shown in Figure 3.

Figure 3 Anatomical basis of lymphatic dissemination in early gastric cancer. A: Distribution of perigastric lymph nodes; B: Distribution of lymphatic vessels in the gastric wall.

Tumor-associated lymphangiogenesis is another critical biological step in the LNM of EGC. It involves proliferation, sprouting, migration, and tubulogenesis of lymphatic endothelial cells (LECs), with proliferation and migration playing particularly important roles[14,15]. Researchers have used specific indicators, such as vascular endothelial growth factor (VEGF) receptor (VEGFR)-3, sex-determining region Y gene-related human menopausal gonadotropin-box 18, and lymphatic vessel endothelial hyaluronan receptor 1, to identify lymphatic structures associated with malignant tumors[15]. VEGF-C and VEGF-D promote LEC proliferation and migration These growth factors are typically produced by various cells in the TME, including tumor cells, tumor-associated fibroblasts, and immune cells. The binding of VEGF-C or VEGF-D to VEGFR-3 activates the VEGF-C/D VEGFR-3 signaling axis. Along this axis, signaling pathways such as PI3K/Akt, TGF-β, and Janus kinase-signal transducer and activator of transcription are activated, which influence the development and alteration of lymphatic vessels in the gastric wall[17,18].

The anatomical distribution of gastric lymphatics and tumor-induced lymphangiogenesis together provide a biological basis for early lymphatic spread in EGC. Abundant submucosal lymphatic channels may allow tumor cells to enter the lymphatic system even when the lesion is still at an early stage. Simultaneously, VEGF-C/D VEGFR-3 signaling promotes lymphatic remodeling around the tumor. However, lymphangiogenesis primarily creates the structural conditions required for lymphatic dissemination, but this does not independently determine the occurrence of LNM. Therefore, LNM risk assessment should also consider the status of the local TME.

TME and immune-mediated lymphatic spread

Regulatory T cells (Tregs) are a specialized group of T cells with immunosuppressive properties. In gastric cancer, these cells are enriched in the TME and draining lymph nodes. Kumar et al[19] reported that Tregs were more abundant in gastric cancer tissues than in healthy tissues, linking Treg accumulation with tumor progression and LNM. Owing to their functional differences, Tregs are divided into natural Tregs (nTregs) and induced Tregs (iTregs)[20]. Compared with thymus-derived nTregs, iTregs are more closely related to the tumor setting because they can develop under cytokine signals such as interleukin (IL)-2 and TGF-β and promote immune escape in gastric cancer[21,22]. FOXP3 supports the suppressive phenotype and functional stability of Tregs and is widely used to detect Treg infiltration in tissue specimens[23]. Immunohistochemical analysis by Zhou et al[24] further showed that FOXP3+ Treg infiltration was significantly associated with LNM (P = 0.024), supporting a role for Treg-mediated immune regulation in lymphatic spread.

Tregs actively interact with LECs, and this intercellular dialogue promotes early LNM[25]. LECs are rich in C-C chemokine ligand 21, which attracts C-C chemokine receptor 7-positive Tregs to the lymphatic microenvironment around tumors. The membrane-bound lymphotoxin (LT) α1β2 on the surface of Tregs binds to LTβR on LECs and activates the nuclear factor kappa-light-chain-enhancer of activated B cells signaling pathway. This interaction changes the structure of lymphatic vessels and increases the motility of both Tregs and cancer cells[26]. In addition, Tregs accumulate around lymphatic vessels and form an immunosuppressive microenvironment with mature regulatory dendritic cells. This environment prevents antigen transport to lymph nodes[27].

The TME is involved in the development of gastric cancer and early LNM[28]. Tumor-associated macrophages (TAMs) arising from peripheral monocytes polarize in response to various cytokines and signaling pathways. M1-type macrophages exhibit antitumor effects, while M2-type macrophages play a tumor-promoting role, including promoting proliferation, angiogenesis, immunosuppression, and metastasis[29]. In EGC, TAMs mainly exhibit an M2-like phenotype, and their number correlates with LNM, indicating their involvement in early tumor spread[30]. M2-type TAMs promote the invasive and migratory behavior of gastric cancer cells since they trigger EMT by releasing matrix metallopeptidase (MMP)-2, MMP-9, and C-X-C CL8[30-33]. Furthermore, they produce lymphangiogenic growth factors, such as VEGF-C and VEGF-D, that stimulate VEGFR-3 signaling in LECs. This interaction leads to lymphangiogenesis and increased permeability, which facilitate malignant cell infiltration into the lymphatic system[34]. TAM-derived cytokines IL-10 and TGF-β are associated with enhanced VEGF-C expression, lymphatic remodeling, and angiogenesis[35-37]. Immunohistochemical analyses have confirmed that high M2-type TAM infiltration correlates with increased lymphatic microvessel density, supporting their contribution to tumor-induced lymphangiogenesis and lymphatic metastasis[38].

The immune and lymphatic pathways have potential therapeutic implications. Potential approaches involve the blockade of LTα1β2-LTβR signaling to disrupt Treg-LEC interactions, inhibition of VEGF-C/D VEGFR-3 signaling to suppress lymphangiogenesis, and targeting of colony-stimulating factor-1/colony-stimulating factor-1 receptor to limit TAM recruitment and polarization. The recently identified angiopoietin-tyrosine kinase with immunoglobulin and epidermal growth factor homology domains-2 and neuropilin-2 pathways have also been explored as potential therapeutic targets, although their clinical validation remains limited[26,39]. EMT-related pathways can also help reduce tumor cell invasiveness, and their integration with strategies targeting the TME may further improve LNM control in EGC[10]. These approaches remain under investigation, and their clinical value in preventing early lymphatic dissemination requires further validation.

These findings suggest that LNM in EGC arises from the coordinated interactions of multiple biological processes. EMT facilitates tumor cell invasion and entry into the lymphatic system. The anatomical distribution of lymphatic vessels within the gastric wall, together with tumor-induced lymphangiogenesis, supports early dissemination. The immune microenvironment influences tumor cell survival and further spread after lymphatic entry. This complexity underlies the challenges in accurate preoperative assessment in EGC.

PREOPERATIVE ASSESSMENT OF LNM RISK

Lymph node status directly determines treatment selection for EGC. LNM in EGC is often small and lacks obvious morphological features, making preoperative detection unreliable in some patients. Imaging, sentinel lymph node (SLN) mapping, molecular assessment, and AI-based models each provide some evidence, but none can define the nodal status with certainty.

Imaging

Computed tomography: Computed tomography (CT) is the principal imaging technique for the evaluation of gastric cancer, both for clinical staging and assessment of nodal status. Multi-detector CT (MDCT) is the most commonly used noninvasive modality for preoperative staging and lymph node evaluation in EGC[40]. MDCT has been reported to achieve an accuracy of approximately 83% for nodal staging, with a sensitivity of 34% and specificity of 91%[41]. However, the diagnostic sensitivity varies across lymph node stations. Xu et al[42] reported that among stations 1-8, station 3 had the highest detection rate of 81%, whereas station 5 had the lowest detection rate of 47%. In clinical practice, MDCT contributes to the detection of LNM and facilitates preoperative N staging of gastric cancer[43].

MDCT is commonly used before treatment because it is accessible and can show the anatomical distribution of lymph nodes. The difficulty is that nodal assessment on MDCT still depends on morphological criteria, such as a short-axis diameter of ≥ 8-10 mm and enhancement pattern. In EGC, this approach is prone to error. Micrometastases may remain invisible in normal-sized nodes, whereas reactive or inflammatory enlargement may be overcalled as metastasis. The sensitivity of MDCT for detecting LNM in EGC is approximately 50%-70%. Therefore, MDCT can support preoperative risk assessment, but a negative result cannot reliably rule out LNM. In addition, inflammatory changes may further reduce diagnostic accuracy and should be carefully considered during preoperative assessment[44-46].

Endoscopic ultrasound: Endoscopic ultrasound (EUS) combines endoscopy with a high-frequency ultrasound probe to visualize the layered structure of the gastric wall and surrounding lymph nodes. Owing to its high spatial resolution, the five-layer structure of the gastric wall and the size and morphology of lymph nodes can be assessed in detail[47]. It is also useful for assessing the depth of tumor invasion in the upper gastrointestinal tract. According to Chen et al[48], the overall accuracy is 78% when compared to histopathological findings, providing clinically useful information for treatment planning and prognostic assessment. EUS has a sensitivity of approximately 60% and specificity of 80% for detecting LNM[49]. The clinical value of EUS in treatment decision-making for EGC is due to its relatively high negative predictive value.

In the absence of suspicious lymph nodes, the likelihood of nodal metastasis may be considered relatively low, which may support the selection of minimally invasive procedures such as ESD. In addition, real-time EUS-guided fine-needle aspiration or biopsy further improves the diagnostic performance. If a suspicious lymph node is identified, EUS can be used to guide the needle through the gastric wall to the target site and obtain a sample for cytological or histological examination. Subsequent pathological evaluation provides pathological confirmation and may further improve the diagnostic accuracy[50].

EUS has certain limitations for the assessment of LNM. Its sensitivity is relatively low and small lymph nodes or metastases with atypical features may be missed, potentially leading to false-negative results. In addition, diagnostic accuracy is influenced by operator experience, resulting in variability in staging assessment. Available studies indicate that EUS is more accurate for T staging than for N staging in gastric cancer and has greater value for prognosis assessment and treatment selection[48]. Therefore, in clinical practice, EUS is more often used to assess the depth of tumor invasion than for independent lymph node assessment.

Positron emission tomography-CT and fibroblast activation protein inhibitor positron emission tomography: Positron emission tomography (PET)-CT is often used for the preoperative assessment of malignant tumors, including gastric cancer. Fluorine-18 fluorodeoxyglucose (¹⁸F-FDG) PET/CT has been increasingly used for the staging and recurrence monitoring of EGC[51]. PET/CT provides both metabolic and anatomical information, allowing for a more comprehensive assessment of regional lymph node involvement[52]. The specificity of PET-CT for detecting lymph node metastases is relatively high at approximately 85%-95%, and metabolically active metastatic lymph nodes can be identified with satisfactory accuracy. Therefore, PET/CT may help to determine the extent of lymph node dissection[53,54]. In EGC, metastatic lymph nodes are often small or limited to micrometastases, and ¹⁸F-FDG uptake by the primary tumor is usually low to moderate. Consequently, the sensitivity of PET-CT for detecting lymph node metastases is significantly lower than that of CT and EUS (approximately 40%), making false-negative results more common in early-stage diseases. Inflamed or reactively enlarged lymph nodes may also show increased FDG uptake, leading to false-positive results and further reducing diagnostic accuracy[52-54].

PET imaging with fibroblast activation protein inhibitor (FAPI) may represent an alternative to ¹⁸F-FDG PET in the assessment of gastric malignancies, especially diffuse and signet-ring cell carcinomas. It targets tumor-associated fibroblasts in the stromal microenvironment of the tumor. Advantages include lower background uptake in the gastrointestinal tract and clearer visualization of lesions, making primary tumors and other metastatic lesions easier to detect[55]. However, FAPI imaging also has limitations. Tracer uptake is not specific to malignant tissues and can occur in benign fibrotic or inflammatory processes, potentially leading to false-positive findings. In addition, FAPI uptake may be low in early-stage tumors, making this method less suitable for early disease detection[56].

In clinical practice, the combination of these modalities improves the assessment of lymph node status; however, limitations remain in the detection of micrometastatic diseases. Therefore, additional approaches, including SLN biopsy, molecular biomarkers, and AI, are being explored to further improve the diagnostic accuracy. Figure 4 summarizes the comparative strengths and limitations of these modalities.

Figure 4 Comparative overview of imaging modalities for lymph node metastasis assessment in early gastric cancer. Computed tomography, endoscopic ultrasound, and positron emission tomography-computed tomography provide complementary anatomical, morphological, and metabolic information, respectively. These modalities have distinct advantages and limitations and collectively contribute to preoperative evaluation of lymph node metastasis in early gastric cancer. PET-CT: Positron emission tomography-computed tomography.

SLN mapping for LNM assessment

In EGC, cancer cells typically metastasize to one or more SLNs before other lymph nodes. This forms the rationale for SLN mapping during surgery to detect LNM. The pathological status of these lymph nodes is used as a predictor of LNM[57]. Through the precise detection and evaluation of SLNs intraoperatively, patients without LNM can be identified and radical lymph node dissection can be avoided. This decreases surgical morbidity, preserves gastric function and enhances quality of life. SLN mapping is usually performed with a dye tracer, a radionuclide, or a combination of both[58,59]. Isosulfan blue dye and technetium-99 m-labeled colloid alone have low detection rates and high false-negative rates. The combined approach increases the detection rate, decreases the false-negative rate, and increases diagnostic accuracy[58-60]. Indocyanine green (ICG) fluorescence imaging is increasingly used for SLN mapping during surgery for EGC because it has high tissue penetration and can be performed in real-time[61]. Furthermore, ICG fluorescence imaging has a high detection rate and is non-radioactive[62,63].

However, some issues are associated with SLN mapping in clinical practice. One of these problems is the complexity of gastric lymphatic drainage, which is characterized by multidirectional flow patterns that can cause unpredictable nodal metastases and may result in skip metastases[64]. The lymphatic drainage of proximal gastric cancers is more variable and unpredictable, which may lead to a higher risk of skip metastasis and false-negative SLN mapping. Distal anterior cancers tend to exhibit more predictable lymphatic drainage patterns. This may enhance the accuracy of SLN mapping[64-67]. The intraoperative pathology is challenging: Conventional frozen section examination and rapid immunohistochemistry have low sensitivity for detecting micrometastases, and section thickness can also affect the accuracy[68]. Furthermore, there are no established guidelines for the tracer type, injection site, timing of injection, or SLN identification, and the accuracy of SLN mapping is highly dependent on the surgeon’s experience, which increases the outcome variability.

At present, SLN mapping is more suitable for patients with EGC treated at high-volume hospitals, where the tumor site is relatively clear and lymph node drainage is predictable. This approach can be used to determine the extent of lymph node dissection, minimize surgical trauma, and enhance quality of life. To facilitate its use, indications for preoperative use must be defined, intraoperative mapping and procedures must be standardized, and the sensitivity of micrometastasis detection must be improved. In the future, SLN mapping alone should not be used to determine the extent of surgery. It should be interpreted together with preoperative imaging, biological features of the tumor, and AI to assess the metastatic risk.

Biomarkers and multi-omics for LNM prediction

When imaging techniques are unable to detect small metastases in EGC, tissue and blood markers derived from the tumor offer additional means of evaluating and predicting LNM. Serum markers are commonly used in clinical practice because they can be easily detected in the blood. Traditional markers, including carcinoembryonic antigen, cancer antigen (CA) 19-9, and CA72-4, are commonly used in the clinical assessment of gastric cancer; however, they have limited value in predicting LNM in EGC. This is largely because they are not sensitive to micrometastases and do not reflect the TME. In a study by Feng et al[69], the positivity rates of these markers were low in patients with early-stage gastric cancer, and although CA19-9 was related to LNM, its sensitivity was low. The pepsinogen I/II ratio, which is an indicator of the status of the gastric mucosa, is associated with the degree of mucosal damage and risk of gastric cancer, but it does not directly indicate metastatic spread[70]. Recently, novel markers such as serum autoantibodies have been explored for early detection and classification, but it is unclear how effective they are in predicting LNM[71]. Since these markers indicate the tumor burden, but not the metastatic potential, they cannot be used in early-stage disease.

The micro-ribonucleic acids have unique expression patterns in gastric cancer patients, which indicates that they can be used as biomarkers to predict LNM[72]. The research on biomarkers is also moving towards tissue markers as opposed to serum markers. In EGC, E-cadherin, suppressor of mother against decapentaplegic homolog 2/3, and VEGF-C/VEGF-D expression have been associated with tumor invasion and LNM. These markers reflect different parts of lymphatic spread, including EMT and lymphangiogenesis[73,74]. An alternative solution to the shortcomings of conventional biomarkers is multi-omics approaches. Transcriptomic and proteomic data have the potential to capture more of the complexity of tumors and can address limitations that cannot be overcome using single biomarkers[75]. These models have the ability to forecast the risk of metastasis more accurately by taking into account not only the alterations in the tumor, but also the properties of the TME, such as the presence of immune cells and the activation of stromal cells[76]. Additionally, they are able to characterize the intricate molecular network of tumor progression and immune remodeling, which gives hints to the mechanisms of metastasis[77].

Multi-omics may enhance the prediction of LNM and the detection of micrometastatic disease. But they are not yet widely used. There are no established data processing protocols or limited clinical validations. Currently, multi-omics is mainly explored in combination with imaging. It can be particularly valuable in patients with equivocal imaging results, suspected micrometastatic disease, or no obvious nodal disease. Multi-omics can help in risk stratification and guide the treatment of EGC when used in conjunction with imaging.

AI in LNM risk prediction

Conventional methods for diagnosing LNM in EGC are limited by their ability to detect small metastatic lesions. AI has demonstrated promising results in imaging and pathology for the noninvasive and accurate prediction of LNM in EGC. Through deep learning and multimodal data fusion, AI models can detect subtle patterns that are difficult to capture with traditional approaches[78,79].

AI models used in preoperative CT image analysis identify imaging features that reflect subtle morphological changes that may not be visible to the human eye, which improves LNM prediction in EGC. He et al[80] developed a deep learning model which integrated 2.5-dimensional radiomics and a multi-instance learning strategy to identify patients with T1N0 staging using preoperative CT scans. The performance of the model was superior to that of traditional radiological interpretations and models that relied on clinical parameters, indicating enhanced diagnostic performance and possible clinical use. Similarly, Hayashi et al[81] discovered that extreme gradient boosting models have the potential to identify patients with pT1b EGC with a low probability of developing LNM. Endoscopy is another potential area for AI. White-light endoscopic images and videos can be used to predict the preoperative and postoperative pathological features of EGC using AI models. These models have favorable predictive accuracies for tumor differentiation and depth of invasion of 89% and 88%, respectively[82].

However, the clinical adoption of AI models has several limitations. Many models are trained on data from one center or a few hospitals, which limits their performance in external patient groups[83,84]. The variability in imaging methods, hardware, and data preprocessing can cause a considerable effect on the performance of the models, which can reduce model accuracy[85]. These problems are even increased in resource-constrained environments, in which the access to cutting-edge imaging, computing resources and clinical information is restricted[86]. In the absence of stable data collection and model development pipelines the variability in performance can be larger, which can cause health care inequities.

Many deep learning models remain “black boxes”, creating both clinical and ethical concerns. Limited interpretability makes clinicians less able to judge why an AI model predicts high LNM risk, which may reduce trust in clinical decision making[87]. To address this issue, model-agnostic explainability methods, such as gradient-weighted class activation mapping, SHapley Additive exPlanations, and local interpretable model-agnostic explanations, have been suggested. These approaches highlight the features or regions used by the model to make predictions and offer more interpretable explanations. However, these methods may not capture the inner workings of the model[88]. Ethical and legal issues, such as liability for errors and when model predictions conflict with clinician intuition, remain unaddressed, and resource allocation may further widen health inequalities[89-91].

AI has the potential to enhance LNM prediction in EGC; however, technical and ethical issues remain and its current use should be seen as adjunctive. In a clinical setting, AI may help in patients with equivocal imaging, which can help more accurately stratify risk and guide decisions regarding endoscopic resection vs surgery. It can help incorporate imaging, pathological, and molecular information to capture aspects that are not captured by conventional assessments.

At present, no single method can completely exclude LNM in EGC because lymphatic spread results from complex biological processes. Table 1 summarizes the major approaches used for LNM assessment in EGC. The key to the future is to combine multimodal information to more accurately stratify the risks and inform treatment decisions for EGC.

Table 1 Comparison of major approaches for lymph node metastasis assessment in early gastric cancer.

Approach	Detection target	Advantages	Limitations	Applicable patients	Current level of evidence
CT	Suspicious lymph nodes and distant disease before treatment	Widely available; useful for routine preoperative staging and assessment	Limited sensitivity for small or microscopic metastases; mainly depends on nodal size and morphology	Patients undergoing routine preoperative staging, especially those with suspicious nodes or high-risk features	Widely used, but limited in detecting micrometastatic LNM in EGC
EUS	Depth of gastric wall invasion and morphology of surrounding lymph nodes	Provides high-resolution visualization of gastric wall layers; useful for estimating invasion depth; EUS-guided biopsy may provide pathological confirmation in selected cases	Operator-dependent; more accurate for T staging than N staging; small metastatic nodes may be missed	Patients requiring pretreatment assessment of invasion depth before ESD or further evaluation of suspicious lymph nodes	Mainly used for depth assessment; provides supportive evidence for LNM assessment
PET-CT/FAPI PET	Metabolically active nodal or distant disease	Provides metabolic and anatomical information	Low sensitivity for EGC and micrometastatic disease; inflammatory or fibrotic uptake may cause false-positive findings	Selected patients with suspected advanced disease, distant metastasis, or inconclusive conventional imaging	PET-CT has selective value; FAPI PET remains promising but insufficiently validated in EGC
Sentinel lymph node mapping	Sentinel lymph nodes for pathological assessment of LNM	May identify patients without nodal metastasis; may reduce unnecessary lymph node dissection; ICG enables real-time visualization	Gastric lymphatic drainage is complex; false-negative results may occur; procedures are not fully standardized	Selected EGC patients in experienced or high-volume centers	Supported in selected settings, but not yet a universally standardized routine pathway
Molecular biomarkers/multi-omics	Serum, tissue, genomic, transcriptomic, proteomic, immune, or stromal features associated with LNM	May provide biological information beyond imaging and reflect EMT, lymphangiogenesis or immune activity	Serum markers have low sensitivity in EGC; omics methods lack standardized protocols and sufficient clinical validation	EGC patients in exploratory or research settings	Exploratory; most evidence remains retrospective or early-stage
AI models	AI-assisted prediction of LNM based on clinical, imaging, or pathological data	May detect subtle patterns that are difficult to identify by conventional assessment	Often based on single-center or small datasets; limited interpretability and external validation	EGC patients in exploratory or adjunctive research settings	Promising in selected studies, but insufficient for independent routine clinical decision-making

CT: Computed tomography; EUS: Endoscopic ultrasound; PET-CT: Positron emission tomography-computed tomography; FAPI: Fibroblast activation protein inhibitor; AI: Artificial intelligence; LNM: Lymph node metastasis; ICG: Indocyanine green; EMT: Epithelial-mesenchymal transition; EGC: Early gastric cancer; ESD: Endoscopic submucosal dissection; PET: Positron emission tomography.

TREATMENT SELECTION BASED ON LNM RISK

In clinical practice, available preoperative evidence can be used to classify patients into low-, intermediate-, and high-risk groups for LNM. Patients with favorable histology and no evidence of nodal involvement are usually suitable for endoscopic resection. Surgery is usually required when deep invasion or suspected LNM is identified on imaging. For patients who fall between standard endoscopic resection and radical surgery, laparoscopic and endoscopic cooperative surgery (LECS) can be considered. This approach allows precise tumor resection with limited surgical burden.

Endoscopic resection for low-risk EGC

For EGC with a low risk of LNM and no evidence of distant disease, endoscopic treatment is the preferred curative approach, with ESD considered the standard of care[3]. According to the 2021 Japanese Gastric Cancer Treatment Guidelines, the absolute indication for endoscopic mucosal resection (EMR) or ESD is limited to differentiated-type adenocarcinoma without ulcerative findings (UL0), clinically diagnosed as T1a, with a lesion diameter ≤ 2 cm. In addition, the guidelines define several absolute indications for ESD, including: (1) Differentiated-type adenocarcinoma UL0, clinically diagnosed as T1a, with a lesion diameter of > 2 cm; (2) Differentiated-type adenocarcinoma with ulcerative findings, clinically diagnosed as T1a, with a lesion diameter ≤ 3 cm; and (3) Undifferentiated-type adenocarcinoma UL0, clinically diagnosed as T1a, with a lesion diameter ≤ 2 cm. Expanded indications include locally recurrent differentiated-type lesions after initial endoscopic resection, categorized as endoscopic curability C-1 for lesions that originally meet the absolute indication, provided that the recurrent lesion is clinically diagnosed as T1a[92,93]. Table 2 summarizes the absolute indications for endoscopic resection of EGC. The procedure involves the submucosal injection of a solution (e.g., normal saline or sodium hyaluronate), followed by en bloc resection using specialized electrosurgical knives. Compared with EMR, ESD allows for the resection of larger lesions (typically > 2 cm) and those with ulceration or irregular morphology. It improves en bloc resection rates by > 90% and enables complete lesion removal[92].

Table 2 Absolute indications for endoscopic resection of early gastric cancer according to the Japanese Gastric Cancer Treatment Guidelines 2021.

Indication	Histology	Depth	Ulcer status	Tumor size	Treatment
Absolute indication	Differentiated-type adenocarcinoma	Mucosa (cT1a)	UL0	≤ 2 cm	EMR or ESD
Absolute indication	Differentiated-type adenocarcinoma	Mucosa (cT1a)	UL0	> 2 cm	ESD
Absolute indication	Differentiated-type adenocarcinoma	Mucosa (cT1a)	UL1	≤ 3 cm	ESD
Absolute indication	Undifferentiated-type adenocarcinoma	Mucosa (cT1a)	UL0	≤ 2 cm	ESD

ESD: Endoscopic submucosal dissection; EMR: Endoscopic mucosal resection; UL0: Without ulceration; UL1: With ulceration.

The success of ESD depends not only on the procedure itself but also on postoperative pathological assessment, as further treatment is based on these results. When assessing the resected specimens, the status of the resection margins, presence or absence of lymphovascular invasion, histological type, tumor size, presence or absence of ulceration, and depth of invasion are evaluated according to the guidelines. These findings determine whether curative resection has been achieved. When the criteria are fulfilled, local tumor control is considered adequate and the risk of LNM is low[94]. By contrast, positive resection margins, deep submucosal invasion, or vascular invasion indicate a higher risk of LNM and may require additional surgical treatment[92,95]. Therefore, postoperative pathological assessment is a decisive factor in determining the subsequent treatment strategy. It directly affects the patient’s prognosis and long-term outcome.

Gastrectomy and lymphadenectomy for high-risk EGC

In cases of EGC with a high risk of LNM, particularly in patients with noncurative ESD or preoperative diagnosis of nodal involvement, extensive surgical resection is an important treatment option[96]. Surgical approaches have evolved with advances in diagnostic imaging, endoscopic technology, and pathological assessment. Treatment is increasingly based on individualized planning in which risk assessment, organ preservation, and minimal invasiveness are key considerations[97].

According to the Japanese guidelines for the management of gastric cancer, the extent of lymph node dissection is determined by the disease stage and the likelihood of nodal metastasis. In early-stage disease, classified as cT1N0, D1 or D1+ lymph node dissection is used, whereas D2 dissection is recommended when nodal involvement is clinically present or in more advanced stages. Surgical treatment of EGC is associated with favorable long-term outcomes and high five-year overall survival rates[92,98].

Laparoscopic approaches have become important surgical options for EGC. Lu et al[99] reported that laparoscopic distal gastrectomy (LADG) provides oncological control comparable to that of open distal gastrectomy. LADG is associated with fewer postoperative complications and faster postoperative recovery. It has become the preferred approach for radical surgery for EGC[99]. Compared with conventional laparoscopy, robotic systems may be advantageous for technically demanding procedures. Their greater instrument stability and three-dimensional imaging may improve surgical accuracy and allow for more precise lymph node dissection[100].

LECS for intermediate or uncertain LNM risk

For patients with intermediate or uncertain risk of LNM, conventional treatment strategies may not provide the most appropriate balance between oncological safety and procedural invasiveness. In this context, LECS has been explored as an alternative approach. LECS combines the minimally invasive advantages of laparoscopy with the precision of endoscopic resection. A single procedure allows for the accurate removal of the gastric lesion, together with targeted lymph node dissection when needed[101]. LECS has been modified to reduce tumor exposure during full-thickness resection. Non-exposed LECS and non-exposed endoscopic wall inversion surgery were developed for this purpose. These techniques help prevent tumor contact with the peritoneal cavity and reduce the risk of intra-abdominal contamination or tumor seeding[102-104].

The combined application of laparoscopic and endoscopic techniques maximizes their complementary strength. LECS reduces the risk of intraoperative bleeding and provides a more refined and minimally invasive treatment option[102-104]. However, several limitations restrict the widespread adoption of LECS. Preoperative assessment of the invasion depth and lymph node status in EGC remains difficult. The results depend partly on the operator experience and may influence the choice of surgical procedure. Furthermore, LECS is technically demanding in complex cases because the operative field is limited. Additionally, internationally standardized protocols for LECS have not yet been established, which restricts its wider clinical use. Figure 5 illustrates the LECS procedure.

Figure 5 Laparoscopic-assisted endoscopic resection. The endoscopic surgeon performs submucosal resection of early gastric cancer, while the laparoscopic team provides external assistance to optimize surgical exposure.

DISCUSSION

LNM risk assessment for EGC still lacks a unified standard. Although tumor biology and diagnostic techniques have advanced, clinicians struggle to translate this information into clear treatment decisions. Current assessments largely rely on physicians’ experience in interpreting diverse, and sometimes conflicting, information. Thus, individualized treatment is difficult.

Early lymphatic spread at the biological level results from interactions between EMT, tumor-induced lymphangiogenesis, and immune interactions in the TME[15,105]. EMT is not merely a change from an epithelial to a mesenchymal phenotype; some tumor cells remain in an intermediate epithelial-mesenchymal state and gain more invasive and metastatic potential[6]. Signals from various cells also control tumor-induced lymphangiogenesis[17,18]. Immune cells in the TME relate tumor invasion to lymphatic remodeling and immune suppression, M2-like macrophages can enhance EMT and are involved in lymphatic remodeling via VEGF-C/D, and VEGF-C/D interacts with Tregs and assists in creating an immune environment that supports metastasis[24,25,30]. These interactions demonstrate that LNM in EGC is a complex biological process. No single mechanism can explain the differences in the risk of nodal metastasis among patients with similar clinicopathological features. The same issue is evident in clinical assessment; using a single imaging characteristic or a single molecular marker, clinicians are unable to obtain an accurate evaluation of the lymph node status.

Although diagnostic technologies are constantly being developed, a single method cannot be used to diagnose LNM in EGC. In daily practice, however, conventional imaging is still the first information that clinicians usually turn to before treatment. Traditional imaging primarily approximates the risk of metastasis based on the size and morphology of lymph nodes. Consequently, small metastatic lesions are often missed[40,47,51]. Visualization of some lesions can be enhanced with newer methods such as FAPI-PET; however, positive results can also be caused by inflammation or fibrosis[55]. Therefore, imaging is useful for routine screening and for detecting clearly suspicious disease, but a negative result should be interpreted cautiously when the pathological or endoscopic features are close to the boundary between ESD and surgery.

The limitation of imaging is not just poor resolution; a single imaging outcome cannot be used to obtain all data required to determine the risk of LNM. This issue has compelled molecular evaluation studies. Multi-omics analysis combines genomic, transcriptomic, and proteomic data, thus providing a more detailed insight into the characteristics of tumors. For borderline EGC, such information may provide additional details that are not captured by conventional imaging or routine pathological assessment. However, this approach remains underutilized in routine clinical application as data processing procedures differ among studies and it is not easy to establish generally acceptable standards[75,76].

SLN mapping attempts to resolve this ambiguity at a procedural level. Its main clinical value lies in borderline cases, where additional nodal information may influence the choice between endoscopic treatment and radical gastrectomy. However, it is highly affected by the location of the tumor and the lymphatic anatomy. This issue is more frequent in tumors and lesions located in the proximity of the lesser curvature, because lymphatic drainage at these locations may take multiple directions rather than a predetermined path of nodal dissemination[64].

Thus, AI models offer a novel integrative method for LNM risk assessment. Unlike conventional imaging or molecular testing, AI can combine imaging, endoscopic, pathological, and molecular information. This gives AI a new role in LNM assessment in EGC. The usefulness of AI is that it can combine preoperative data and transform it into a larger estimate of LNM risk[78,79]. This may be particularly useful for borderline EGC, where the choice between ESD and surgery is often made with incomplete information. Nonetheless, there are obvious limitations to existing research. Most AI models are trained using single-center data; therefore, their training data may be limited to one institution or a small patient population. Thus, their performance in the initial cohort does not imply the same accuracy in other cohorts. Another limitation of deep learning models in clinical use is their poor interpretability. Thus, the risk score is insufficient for preoperative treatment selection, and clinicians should understand the cause of risk estimates[87,88,90]. In current clinical practice, AI can be more helpful as an aid in structuring multisource information and assisting in risk stratification; it is too unreliable as a tool to independently make a decision on whether a patient should undergo ESD or radical surgery.

The most challenging treatment decisions are those for patients with borderline or uncertain risks. ESD is generally used to treat low-risk lesions to achieve local control without compromising gastric function[92,97]. High-risk patients frequently require radical gastrectomy with lymph node dissection to minimize the risk of persisting metastatic disease[92]. The difficult group lies between these two situations. A typical example is a patient who undergoes ESD and is then found to have superficial submucosal invasion but no lymphovascular invasion. In some guidelines and centers, such lesions may fall close to the expanded indications for endoscopic treatment, but the decision to observe or perform additional surgery is still not uniform in clinical practice. In patients with moderate or unclear risk, a trade-off between overtreatment and undertreatment is required in the choice of treatment. LECS can strike a balance between local resection and functional preservation in selected cases and offers a more flexible treatment choice for such patients[101-104]. This renders proper preoperative evaluation of the LNM risk key to maximizing treatment plans for EGC.

CONCLUSION

Preoperative prediction of LNM is still difficult in EGC because lymphatic spread involves complex biological processes and current diagnostic methods have clear limitations. Further studies are needed to combine clinicopathological, imaging, and molecular data to create a more integrated risk-scoring system and assessment pathway. This integration would offer a more solid foundation of individualized decisions between endoscopic resection and surgery in EGC patients.