BPG is committed to discovery and dissemination of knowledge
Retrospective Study Open Access
Copyright: ©Author(s) 2026. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) license. No commercial re-use. See permissions. Published by Baishideng Publishing Group Inc.
World J Gastrointest Oncol. Nov 15, 2026; 18(11): 123321
Published online Nov 15, 2026. doi: 10.4251/wjgo.123321
Preoperative dual-energy computed tomography radiomics for prognosis of advanced gastric adenocarcinoma
Ao Liu, Min Li, Yang You, Lei Gao, Xiao-Sheng Xu, Yu Li, Jing Liu, Li Yang, Department of Computed Tomography and Magnetic Resonance, Fourth Hospital of Hebei Medical University, Shijiazhuang 050000, Hebei Province, China
Yong Li, Zhi-Dong Zhang, Third Department of Surgery, Fourth Hospital of Hebei Medical University, Shijiazhuang 050000, Hebei Province, China
ORCID number: Li Yang (0000-0003-1053-4519).
Co-first authors: Ao Liu and Min Li.
Author contributions: Liu A and Li M wrote the first draft of the manuscript as co-first authors; all authors contributed to the study conception and design, and commented on previous versions of the manuscript; You Y, Xu XS, Li Yu, Liu J, Li Yong, and Zhang ZD performed data collection; Liu A, Gao L, and Li M performed data analysis; Yang L reviewed and edited the manuscript. All authors read and approved the final manuscript.
AI contribution statement: Portions of this manuscript were edited using AI tools solely for language refinement. The authors carefully reviewed and verified all AI-assisted outputs and take full responsibility for the scientific content of the manuscript.
Supported by S&T Program of Hebei, No. 22377789D and No. 22377701D; and Medical Science Research Project of Hebei, No. 20221314.
Institutional review board statement: This retrospective study was approved by the Ethics Committee of the Fourth Hospital of Hebei Medical University, No. 2021KY347.
Informed consent statement: Given the retrospective design and use of de-identified patient data, the requirement for obtaining written informed consent was waived by the ethics committee.
Conflict-of-interest statement: All the authors report no relevant conflicts of interest for this article.
Data sharing statement: The data that support the findings of this study are available from the corresponding author upon reasonable request.
Corresponding author: Li Yang, MD, Department of Computed Tomography and Magnetic Resonance, Fourth Hospital of Hebei Medical University, No. 12 Jiankang Road, Chang’an District, Shijiazhuang 050000, Hebei Province, China. 48000664@hebmu.edu.cn
Received: May 15, 2026
Revised: June 25, 2026
Accepted: July 31, 2026
Published online: November 15, 2026
Processing time: 174 Days and 23.9 Hours

Abstract
BACKGROUND

Gastric cancer remains a major global health burden. Many patients are diagnosed at an advanced stage and remain at high risk of recurrence despite curative-intent surgery. Although pathological tumor-node-metastasis (pTNM) staging is the cornerstone of prognostic assessment, patients within the same stage often show heterogeneous outcomes, suggesting that additional tumor biology-related information is needed to refine prognostic assessment. Dual-energy computed tomography (DECT)-derived iodine maps reflect tumor perfusion and vascularity, and radiomics can quantify subtle intratumoral heterogeneity. We hypothesized that iodine-map-based DECT radiomics, integrated with clinical factors, could improve individualized prognostic predictions in resectable advanced gastric adenocarcinoma.

AIM

To develop and validate DECT radiomics-based models for disease-free survival and overall survival prediction in resectable advanced gastric adenocarcinoma.

METHODS

This retrospective study included 175 patients with advanced gastric adenocarcinoma who underwent radical gastrectomy and were divided into training (n = 122) and validation (n = 53) groups. Radiomic features were extracted from venous-phase-blended images and iodine maps. Prognostic models based on pTNM staging, clinical factors, radiomics signature (Rad-score), and combined clinical-radiomics features were developed and validated using Cox regression, Harrell’s C-index, receiver operating characteristic curves, calibration plots, and decision curve analysis.

RESULTS

The clinical-radiomics model, comprising the Rad-score, invasive morphology, elevated serum carcinoembryonic antigen level, and positive clinical N stage, demonstrated superior predictive accuracy compared with the pTNM model for disease-free survival (C-index: Training, 0.771 vs 0.686; validation, 0.712 vs 0.665) and overall survival (C-index: Training, 0.821 vs 0.695; validation, 0.714 vs 0.689). The clinical-radiomics model had a lower Akaike information criterion value than the pTNM model, indicating a better model fit. Subgroup analyses confirmed robust risk stratification of patients with clinically advanced disease.

CONCLUSION

The combined clinical-radiomics model integrating DECT radiomics with key clinical prognostic factors may improve preoperative risk stratification in resectable advanced gastric adenocarcinoma and help guide individualized treatment strategies.

Key Words: Advanced gastric cancer; Dual-energy imaging; Radiomics; Prognosis; Risk stratification; Survival analysis

Core Tip: This study developed a preoperative dual-energy computed tomography radiomics model for disease-free and overall survival prediction in resectable advanced gastric adenocarcinoma. By integrating radiomics features with clinical nodal status, tumor morphology, and carcinoembryonic antigen level, the model improved exploratory risk stratification over pathological tumor-node-metastasis staging.



INTRODUCTION

Gastric cancer (GC) remains a significant global health challenge, ranking as both the fifth most common cancer and a leading cause of cancer-related mortality worldwide[1]. Owing to the often subtle nature of early symptoms, most patients present with advanced-stage disease at initial diagnosis. Consequently, even after undergoing curative-intent radical surgery, these patients have high recurrence rates and modest survival outcomes[2-4]. Advances in medicine have led to a paradigm shift in advanced GC management, moving beyond surgery alone towards comprehensive, surgery-centered approaches[5,6]. Accurate individualized prognostic prediction is essential for enabling personalized precision treatment.

Currently, the postoperative pathological tumor-node-metastasis (pTNM) staging system serves as the cornerstone for the clinical prognostic assessment of patients with GC. However, considerable heterogeneity in clinical outcomes is frequently observed among patients who share the same pTNM stage and receive similar treatment regimens[7-9]. This prognostic variability is likely due to several factors. First, the pTNM system does not encompass other established prognostic indicators, such as Borrmann type, tumor location, histological subtype, or preoperative serum tumor marker levels[10-12]. Second, GC is characterized by substantial biological heterogeneity; The Cancer Genome Atlas classified gastric adenocarcinoma into molecular subtypes with distinct prognoses and treatment responses, while differences in tumor vascularity and immune-stromal microenvironments may further contribute to outcome variability among patients with the same pTNM stage[13-15].

Radiomics, an emerging noninvasive approach that extracts abundant quantitative features from routine medical images, could capture subtle intratumoral variations associated with inherent biological tumor characteristics[16]. In GC, previous studies have shown that computed tomography (CT)-based radiomic signatures can predict overall survival (OS) and disease-free survival (DFS), improve the prognostic value of TNM staging, and support preoperative prediction of early recurrence and treatment response[17-19]. Dual-energy CT (DECT)-derived iodine maps (IM) quantitatively reflect tumor perfusion and vascularity, which are closely associated with tumor biological aggressiveness[20]. Accordingly, we hypothesized that radiomic features from DECT IM could provide more biologically meaningful prognostic information than conventional imaging.

Therefore, in this study, we aimed to develop and validate prognostic models for patients with resectable advanced gastric adenocarcinoma using radiomics features extracted from both venous-phase-blended images and IM obtained from preoperative DECT scans. By integrating these radiomic signatures with key clinical prognostic factors, we predicted DFS and OS to develop robust predictive models for patient risk stratification and guide individualized precision therapy.

MATERIALS AND METHODS
Ethical approval and reporting guidelines

This retrospective study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Fourth Hospital of Hebei Medical University, No. 2021KY347. Given the retrospective design and use of de-identified patient data, the requirement for obtaining written informed consent was waived by the ethics committee. This manuscript has been prepared following the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis guidelines[21], which provide a standardized framework for the development and validation of prediction models.

Study population

We retrospectively collected data from patients who underwent surgical resection for GC at our institution between April 2015 and July 2018. The inclusion criteria were as follows: (1) Patients who underwent radical gastrectomy with curative intent, and whose postoperative pathology confirmed advanced gastric adenocarcinoma (stage pT2-4aN0-3M0, corresponding to stage IB - stage IIIC according to the American Joint Committee on Cancer staging manual, 8th edition); (2) No history of preoperative anti-tumor therapy (such as neoadjuvant chemotherapy or radiotherapy); (3) Availability of a preoperative abdominal dual-source CT examination performed within 14 days prior to surgery; (4) Received standard postoperative adjuvant treatment according to clinical guidelines; and (5) Had complete follow-up information for a minimum of 5 years post-surgery, or until the date of death if it occurred within the 5-year period. Patients were excluded based on the following criteria: (1) Incomplete clinical records, pathological information, and imaging data; and (2) History or concurrent presence of other malignant neoplasms. A flowchart detailing the patient selection process, including the number of patients screened, included, and excluded along with their allocation to the training and validation sets, is presented in Figure 1A.

Figure 1
Figure 1 Study flowchart. A: Acquisition of data and radiomic features: Including patient enrollment, image processing and volume of interest segmentation, and radiomic feature extraction; B: Screening of radiomic and clinical features, and construction of prognostic models; C: Performance evaluation and clinical application of the prognostic model. CT: Computed tomography; VOI: Volume of interest; ICC: Intraclass correlation coefficient; pTNM: Pathological tumor-node-metastasis; CEA: Carcinoembryonic antigen; AUC: Area under the curve; DFS: Disease-free survival.
Patient follow-up and endpoint definitions

Postoperative follow-up was conducted by telephone and clinical record reviews. The present study used a prespecified 5-year follow-up window. Patients were included if they had completed 5-year follow-up or experienced the corresponding endpoint event within 5 years after radical surgery. Follow-up information beyond 5 years was not included in the present analysis. The primary endpoints were OS and DFS. OS and DFS were defined as the interval from surgery to all-cause death or censoring at the final follow-up and the time from surgery to first recurrence, metastasis, or all-cause death, respectively[22,23]. Patients without disease progression at the last follow-up were censored.

CT image acquisition protocol

All patients received standardized preoperative abdominal CT with a second-generation dual-source CT system (SOMATOM Definition Flash, Siemens Healthineers, Germany). Unenhanced and dual-energy contrast-enhanced scans were performed under standardized protocols (for detailed acquisition parameters, see Supplementary material).

CT image feature evaluation

Two blinded abdominal radiologists (observers 1 and 2, 5 and 20 years of experience, respectively) independently evaluated preoperative CT images, masked to pathological and clinical outcomes. Evaluations relied primarily on axial images with auxiliary multiplanar reconstructions, and inter-observer discrepancies were resolved by consensus. Supplementary material contains the detailed definitions and classification criteria for all CT features.

Radiomics workflow

Tumor segmentation was performed by Observer 1, referencing the corresponding coronal and sagittal planes to ensure accurate three-dimensional delineation (Figure 1A). To evaluate inter-observer feature reproducibility, 30 cases were selected using a computer-generated random number method. In those cases, Observer 2 independently performed volume of interest segmentation while blinded to Observer 1’s segmentation results. Radiomic features were extracted after volume of interest delineation using feature definitions that followed the Image Biomarker Standardization Initiative recommendations where applicable[24]. Details regarding the extracted feature categories are provided in the Supplementary material. Only features with an intraclass correlation coefficient (ICC) ≥ 0.80 were retained. To reduce overfitting and selection bias, feature selection was conducted exclusively in the training set. The least absolute shrinkage and selection operator-penalized Cox proportional hazards regression model was applied using DFS as the endpoint to identify predictive radiomic features. The optimal λ parameter was selected using 10-fold cross-validation based on the minimum-error criterion, defined as the λ value corresponding to the minimum mean cross-validated partial likelihood deviance.

Development and validation of prognostic models

The patients were randomly divided into a training set and a validation set at a ratio of 7:3 using R software with a fixed random seed to ensure reproducibility. To predict DFS and OS, four prognostic models were developed using the training set and validated in the validation set based on Cox proportional hazards regression. A pTNM model was constructed using postoperative pT and pN stages. A clinical model was developed by selecting significant baseline clinical and CT variables through univariate and multivariable analyses. The candidate variables included age, sex, body mass index, presence of comorbidities, preoperative serum carcinoembryonic antigen (CEA), carbohydrate antigen 19-9, and carbohydrate antigen 72-4 levels, tumor location, morphology, and thickness, tumor enhancement pattern, degree of tumor enhancement, clinical tumor (cT) stage, and clinical nodal (cN) stage. A radiomics model was built using a radiomics signature (Rad-score) derived from selected features. Finally, a combined clinical-radiomics model integrated independent clinical predictors with the Rad-score.

For risk stratification, patients were categorized into low- and high-risk groups based on optimal cutoff values derived from the combined model. Survival differences between groups were evaluated using Kaplan-Meier analysis and log-rank tests. Additional subgroup analyses were performed to assess model performance across clinically relevant strata. The overall study workflow is illustrated in Figure 1.

Statistical analysis

All statistical analyses were performed using R (version 4.3.1) and Python (version 3.9.12). Baseline characteristics were compared between the training and validation sets using the Wilcoxon rank-sum test for continuous or ordinal variables, and the χ2 or Fisher’s exact test for categorical variables, as appropriate. Cox proportional hazards regression was used for univariate screening and multivariate model construction. Hazard ratios with 95% confidence intervals (CIs) were calculated for DFS and OS. The predictive performance of the pTNM, clinical, radiomics, and clinical-radiomics models was evaluated in both cohorts using Harrell’s C-index, time-dependent receiver operating characteristic curves, calibration curves, decision curve analysis, and Akaike information criterion (AIC). Differences in the C-index between models were assessed using DeLong’s test. Ten-fold cross-validation was performed in the training cohort to assess the model robustness. All tests were two-sided, and statistical significance was set at P < 0.05.

RESULTS
Patient characteristics

A total of 175 patients fulfilled the inclusion criteria and were enrolled in the study (Figure 1A). The cohort consisted primarily of 145 male patients (82.9%) and 30 female patients (17.1%). The median age at surgery was 61 years (range: 27-80 years). During the 5-year follow-up period, disease recurrence or metastasis was observed in 67 patients (38.3%) and 59 patients (33.7%) died from any cause. Within this prespecified 5-year follow-up window, the median follow-up duration was 5 years for the entire cohort, training set, and validation set in both DFS and OS analyses. The patients were randomly allocated to a training set (n = 122) or a validation set (n = 53) in a 7:3 ratio. The baseline clinical characteristics, laboratory findings, radiological features, and pathological details were well balanced between the training and validation sets, with no statistically significant differences observed (all P > 0.05, Supplementary Table 1).

Model development

In the entire cohort (n = 175), the overall distribution according to the postoperative pathological stage (American Joint Committee on Cancer 8th edition) was as follows: Stage IB (n = 11), stage II (n = 41), and stage III (n = 123). The pTNM model was constructed using the pT and pN stage categories as predictors in a predefined multivariate Cox model. In the clinical model, univariate Cox analysis of the training cohort identified preoperative serum CEA and carbohydrate antigen 19-9 levels, tumor morphology, tumor thickness, and clinical N stage as significant predictors of DFS (all P < 0.05). Subsequent multivariate Cox regression analysis using backward elimination retained elevated serum CEA levels, invasive tumor morphology, and positive cN stage as independent prognostic factors (Supplementary Table 2). These three variables were used to construct the final clinical model.

For the radiomics model, 1691 features were initially extracted from each of the blended images and IM. Following assessment of inter-observer reproducibility, 2720 features demonstrating good to excellent stability (ICC ≥ 0.80) were kept (Figure 1B). Least absolute shrinkage and selection operator-Cox regression analysis applied to the training set identified a final signature comprising nine optimal radiomic features with nonzero coefficients (Figure 1B). Notably, five of these nine features were derived from IM, and their mean absolute SHapley Additive exPlanations values were higher than those of the blended image-derived features (Figure 2A-D). A Rad-score was calculated for each patient via a linear combination of these selected features weighted by their respective coefficients (the coefficients are detailed in Supplementary Table 3). The Rad-score forms the basis of the radiomics model. Finally, a clinical-radiomics model was developed by integrating the three independent clinical predictors (invasive morphology, positive cN stage, and elevated CEA) and the calculated Rad-score into a multivariate Cox proportional hazards model.

Figure 2
Figure 2 Feature contribution and Kaplan-Meier risk stratification of prognostic models. A and B: Overall contribution of radiomic features to disease-free survival/overall survival prediction quantified by mean absolute SHapley Additive exPlanations values; C and D: Magnitude and direction of feature impact on model output visualized via SHapley Additive exPlanations summary plots; E-H: Kaplan-Meier curves of high- and low-risk subgroups among all patients in the training and validation sets; I-L: Kaplan-Meier curves of high- and low-risk subgroups among pathological stage III patients in the training and validation sets. DFS: Disease-free survival; OS: Overall survival.

The optimal cutoff values for risk stratification were 1.778 and 1.609 for the DFS and OS risk scores, respectively. Using these thresholds, patients were successfully stratified into low- and high-risk groups, exhibiting significantly different DFS and OS outcomes (all P < 0.05; Figure 2E-H). Importantly, subgroup analyses confirmed that the clinical-radiomics model maintained significant prognostic stratification ability in patients with postoperative pathological stage III (Figure 2I-L) as well as in those with cT4 and cN+ (Supplementary Figures 1 and 2); all P < 0.05. Confusion matrices and concentric circle plots were used to visualize the classification performance and relative clinical applicability of the combined model (Figure 3; Supplementary Figures 1 and 2). To enhance clinical applicability, nomograms based on the clinical-radiomics model were constructed to provide individualized predictions of 1-, 3-, and 5-year DFS and OS probabilities (Figure 4A and B).

Figure 3
Figure 3 Confusion matrices and concentric plots showing the classification performance and clinical applicability of the combined model. A-H: Confusion matrices and concentric plots for disease-free survival and overall survival prediction in all patients and patients with pathological stage III disease in the training set; I-P: Corresponding analyses in the validation set. DFS: Disease-free survival; OS: Overall survival; AIC: Akaike information criterion; TNM: Tumor-node-metastasis; AUC: Area under the curve; CI: Confidence interval.
Figure 4
Figure 4 Nomograms and comprehensive model performance evaluation. A and B: A patient with pT4N3M0 staging in the validation cohort was stratified into the low-risk group by the combined model; the nomogram indicated that the probabilities of the patient having disease-free survival and overall survival of less than 5 years were 24.8% and 14.1%, respectively. Ultimately, the patient remained alive at 5 years postoperatively without recurrence or metastasis; C and D: C-index comparison among models via 10-fold cross-validation in the training cohort; E-H: Decision curve analysis in the training and validation sets; I and J: Akaike information criterion values of each prognostic model in the validation set; K-N: Calibration curves in the training and validation sets; O-R: Receiver operating characteristic curves in the training and validation sets. DFS: Disease-free survival; OS: Overall survival; AIC: Akaike information criterion; TNM: Tumor-node-metastasis; AUC: Area under the curve; CI: Confidence interval.
Evaluation and comparison of model predictive performance

In the training set, the clinical-radiomics model demonstrated superior discrimination ability compared to the other models. To predict DFS, the C-indices were 0.771 (95%CI: 0.713-0.830) for the clinical-radiomics model, 0.686 (95%CI: 0.612-0.759) for the pTNM model, 0.693 (95%CI: 0.625-0.761) for the clinical model, and 0.708 (95%CI: 0.637-0.779) for the radiomics model. Similarly, for OS prediction, the C-indices were 0.821 (95%CI: 0.757-0.884) for the clinical-radiomics model, 0.695 (95%CI: 0.612-0.778) for the pTNM model, 0.696 (95%CI: 0.621-0.771) for the clinical model, and 0.758 (95%CI: 0.683-0.834) for the radiomics model. DeLong’s test confirmed that the C-indices of the clinical-radiomics model were significantly higher than those of the other three models in the training set for both DFS and OS predictions (all P < 0.05; Supplementary Table 4). In the validation set, the clinical-radiomics model maintained the highest C-index values for both DFS (0.712, 95%CI: 0.615-0.809) and OS (0.714, 95%CI: 0.616-0.812). However, the difference in performance between the pTNM (DFS, 0.665; OS, 0.689) and radiomics models (DFS, 0.642; OS, 0.621) was not statistically significant (P > 0.05). The clinical-radiomics model’s C-index for DFS was significantly higher than that of the clinical model (0.677, P = 0.029), whereas the difference in OS prediction (clinical model C-index: 0.688) was not significant (P = 0.142; Supplementary Table 4). Additionally, a 10-fold cross-validation was conducted on the training cohort to verify the robustness of the model performance, which yielded consistent results, and the combined model still had the highest median C-index for DFS and OS among all the models (Figure 4C and D).

Decision curve analysis demonstrated that the clinical-radiomics model offered the highest net clinical benefit across a wide range of threshold probabilities for both DFS and OS prediction compared to the other models in both the training and validation sets (Figure 4E-H). Evaluation using AIC indicated that the clinical-radiomics model generally exhibited a better balance between goodness-of-fit and complexity (lowest AIC values), except when compared with the simpler clinical model in the validation set (Figure 4I and J; Supplementary Table 5).

The calibration curves showed good agreement between the survival probabilities predicted by the clinical-radiomics model and the actual observed survival outcomes, suggesting reliable probability estimates (Figure 4K-N). Furthermore, the time-dependent receiver operating characteristic analysis confirmed the excellent predictive performance of the clinical-radiomics model over the follow-up period (Figure 4O-R).

DISCUSSION

In this study, we developed and validated a model based on dual-source DECT radiomics and clinical prognostic factors to predict the prognosis of patients with resectable advanced gastric adenocarcinoma. This model can preoperatively predict the 1-, 3-, and 5-year DFS and OS of patients after radical resection, allow for prognostic risk stratification of patients, and provide assistance in guiding the selection of optimal clinical treatment strategies and achieving individualized precision therapy.

In the clinical model, tumor morphology, serum CEA level, and cN stage demonstrated independent predictive value for the prognosis of patients with resectable advanced gastric adenocarcinoma, and numerous prior studies have confirmed that these factors are associated with prognosis in GC[25-31]. Although the depth of invasion in GC is also an important prognostic indicator, T stage did not emerge as an independent predictive feature in this study, likely because the majority of patients in this cohort had cT4 disease (140/175, 80.0%). This predominance of locally advanced tumors may have reduced the variability and discriminatory ability of the cT stage, thereby decreasing its statistical weight in the multivariable clinical model. In such a relatively homogeneous advanced-T-stage population, other factors, like the cN stage, tumor morphology, CEA level, or radiomic features, may provide additional prognostic information for distinguishing patients with different risks. Future studies including patients with a more balanced distribution of T stages are needed to better evaluate the independent prognostic contribution of the T stage and improve the generalizability of the clinical-radiomics model.

Tumor segmentation is a critical step in radiomic analysis. We applied three-dimensional tumor segmentation to capture whole-tumor characteristics, ensuring robust and reproducible radiomic feature extraction[32]. The final radiomic signature comprised nine features, five of which were IM-derived. SHAP analysis indicated that these iodine-map features contributed most strongly to the predictions of the model, highlighting the added prognostic value of dual-energy CT-derived iodine data. However, SHAP analysis reflects model-based feature importance rather than direct evidence of biological causality. Although radiomics can capture subtle intratumoral heterogeneity[24,33,34], the biological interpretability of individual features remains limited[35], and further histopathological validation, such as correlating IM radiomic features with microvascular density or angiogenesis-related markers, is needed to confirm their underlying biological significance.

Our clinical-radiomics model integrating radiomic features and clinical predictors demonstrated favorable prognostic performance in both the training and validation cohorts. However, in the validation cohort, the difference in C-index between the clinical-radiomics model and the pTNM model did not reach statistical significance. This may be partly attributable to the relatively small size of the validation cohort, which may have increased the uncertainty of performance estimates and limited the statistical power for model comparison. To further evaluate the model fit, we supplemented the analysis with an AIC comparison and found that the clinical-radiomics model showed a lower AIC than the pTNM model. In addition, cross-validation supported the stability of the clinical-radiomics model. Importantly, the model enabled meaningful risk stratification within the same pathological stage, identifying patients at higher or lower risk of recurrence and death. These findings suggest that combining clinical and radiomic features may provide additional preoperative prognostic information beyond conventional staging, although further validation in larger external cohorts is needed before broader clinical application.

This study has several limitations. First, this was a single-center retrospective study with a relatively small sample size, including only 53 patients in the validation cohort, which may have introduced selection bias, increased the uncertainty of performance estimates, and limited the generalizability of the model. In addition, all CT examinations were performed using a single dual-source DECT scanner, and the reproducibility of radiomic features across different CT scanners, vendors, and acquisition protocols was not evaluated. Therefore, external multicenter validation is required before broader clinical application. Second, the patient cohort predominantly comprised cT4 and cN+ cases, representing later-stage disease. This stage imbalance might have affected the model’s performance. Third, although DECT imaging was performed, spectral features were not included in the model construction to preserve the convenience of the model application. In addition, radiomic feature selection was performed using DFS as the primary endpoint, and the same DFS-derived radiomic signature was subsequently applied to OS prediction, which may have introduced potential endpoint-related optimization bias. Future studies should further evaluate endpoint-specific radiomic signatures for DFS and OS separately. Fourth, the radiomic analysis relied on manual tumor segmentation, which is time-consuming and may be susceptible to inter-observer variability. We performed ICC-based testing to account for reproducibility issues. Future studies may incorporate semi-automatic or automatic segmentation methods to improve workflow efficiency and reproducibility. Fifth, this study focused on routine preoperative variables and imaging features for prognostic prediction. Molecular and pathological biomarkers, such as human epidermal growth factor receptor 2 status, microsatellite instability status, programmed death-ligand 1 expression, and Epstein-Barr virus status[13-15], as well as detailed surgical factors, such as the surgical approach and quantitative lymph node dissection metrics[36,37], and postoperative factors, such as the quality of postoperative recovery and nutritional support[38], were not incorporated into the present model. Incorporating these factors may further improve prognostic accuracy and should be considered in future model development. In addition, patients who received preoperative anti-tumor therapy were excluded to maintain cohort homogeneity. However, because neoadjuvant chemotherapy is increasingly used as a standard treatment option for advanced GC and may alter tumor biology, induce psychological stress, and affect prognosis[39], this exclusion may limit the generalizability of our model to patients receiving neoadjuvant therapy. Future studies should incorporate neoadjuvant treatment status, treatment response, and systemic impacts.

CONCLUSION

Preoperative risk stratification using a model combining dual-energy CT radiomics and key clinical factors (tumor morphology, CEA level, and cN stage) accurately predicted survival outcomes in patients with resectable advanced gastric adenocarcinoma. This tool holds promise for guiding personalized treatment strategies by identifying patient risks before surgery.

References
1.  Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74:229-263.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 16785]  [Cited by in RCA: 18099]  [Article Influence: 9049.5]  [Reference Citation Analysis (55)]
2.  Smyth EC, Nilsson M, Grabsch HI, van Grieken NC, Lordick F. Gastric cancer. Lancet. 2020;396:635-648.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 4042]  [Cited by in RCA: 3679]  [Article Influence: 613.2]  [Reference Citation Analysis (26)]
3.  Li GZ, Doherty GM, Wang J. Surgical Management of Gastric Cancer: A Review. JAMA Surg. 2022;157:446-454.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 230]  [Cited by in RCA: 229]  [Article Influence: 57.3]  [Reference Citation Analysis (9)]
4.  Shang-Guan ZX, Zhong Q, Zhang ZQ, Liu ZY, Sun YQ, Ma YB, Ding FH, Wu SC, Zhu JY, Wang YH, Zhao W, Yu JH, Li BL, Wu J, Ye W, Li P, Xie JW, Zheng CH, Chen QY, Huang CM. Surgical and oncological outcomes of laparoscopic versus open gastrectomy after neoadjuvant chemotherapy in patients with locally advanced gastric cancer: A multicenter analysis. Eur J Surg Oncol. 2025;51:110060.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 3]  [Cited by in RCA: 2]  [Article Influence: 2.0]  [Reference Citation Analysis (0)]
5.  Shi Y, Xu X, Zhao Y, Qian F, Tang B, Hao Y, Luo H, Chen J, Yu P. Long-term oncologic outcomes of a randomized controlled trial comparing laparoscopic versus open gastrectomy with D2 lymph node dissection for advanced gastric cancer. Surgery. 2019;165:1211-1216.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 56]  [Cited by in RCA: 51]  [Article Influence: 7.3]  [Reference Citation Analysis (11)]
6.  Seeneevassen L, Bessède E, Mégraud F, Lehours P, Dubus P, Varon C. Gastric Cancer: Advances in Carcinogenesis Research and New Therapeutic Strategies. Int J Mol Sci. 2021;22:3418.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 23]  [Cited by in RCA: 131]  [Article Influence: 26.2]  [Reference Citation Analysis (4)]
7.  Bang YJ, Kim YW, Yang HK, Chung HC, Park YK, Lee KH, Lee KW, Kim YH, Noh SI, Cho JY, Mok YJ, Kim YH, Ji J, Yeh TS, Button P, Sirzén F, Noh SH; CLASSIC trial investigators. Adjuvant capecitabine and oxaliplatin for gastric cancer after D2 gastrectomy (CLASSIC): a phase 3 open-label, randomised controlled trial. Lancet. 2012;379:315-321.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 1359]  [Cited by in RCA: 1370]  [Article Influence: 97.9]  [Reference Citation Analysis (32)]
8.  Kattan MW, Karpeh MS, Mazumdar M, Brennan MF. Postoperative nomogram for disease-specific survival after an R0 resection for gastric carcinoma. J Clin Oncol. 2003;21:3647-3650.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 346]  [Cited by in RCA: 341]  [Article Influence: 14.8]  [Reference Citation Analysis (3)]
9.  Woo Y, Son T, Song K, Okumura N, Hu Y, Cho GS, Kim JW, Choi SH, Noh SH, Hyung WJ. A Novel Prediction Model of Prognosis After Gastrectomy for Gastric Carcinoma: Development and Validation Using Asian Databases. Ann Surg. 2016;264:114-120.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 33]  [Cited by in RCA: 42]  [Article Influence: 4.2]  [Reference Citation Analysis (3)]
10.  Zhang R, Chen X, Chen G, Zhao Z, Wei Y, Zhang F, Lin J, Nie R, Chen Y. Combined Use of Tumor Markers in Gastric Cancer: A Novel Method with Promising Prognostic Accuracy and Practicality. Ann Surg Oncol. 2023;30:8561-8571.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 15]  [Cited by in RCA: 17]  [Article Influence: 5.7]  [Reference Citation Analysis (3)]
11.  Choi S, Song JH, Lee S, Cho M, Kim YM, Kim HI, Hyung WJ. Lymphovascular Invasion: Traditional but Vital and Sensible Prognostic Factor in Early Gastric Cancer. Ann Surg Oncol. 2021;28:8928-8935.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 1]  [Cited by in RCA: 39]  [Article Influence: 7.8]  [Reference Citation Analysis (0)]
12.  Li P, He HQ, Zhu CM, Ling YH, Hu WM, Zhang XK, Luo RZ, Yun JP, Xie D, Li YF, Cai MY. The prognostic significance of lymphovascular invasion in patients with resectable gastric cancer: a large retrospective study from Southern China. BMC Cancer. 2015;15:370.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 45]  [Cited by in RCA: 57]  [Article Influence: 5.2]  [Reference Citation Analysis (0)]
13.  Cancer Genome Atlas Research Network. Comprehensive molecular characterization of gastric adenocarcinoma. Nature. 2014;513:202-209.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 5694]  [Cited by in RCA: 5313]  [Article Influence: 442.8]  [Reference Citation Analysis (18)]
14.  Feng W, Wang Y, Chen S, Zhu X. Intra-tumoral heterogeneity and immune responses predicts prognosis of gastric cancer. Aging (Albany NY). 2020;12:24333-24344.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 4]  [Cited by in RCA: 9]  [Article Influence: 1.5]  [Reference Citation Analysis (0)]
15.  Hong WG, Ko YS, Pyo JS. Clinicopathological significance and prognostic role of microvessel density in gastric cancer: A meta-analysis. Pathol Res Pract. 2017;213:1459-1463.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 11]  [Cited by in RCA: 12]  [Article Influence: 1.3]  [Reference Citation Analysis (2)]
16.  Avanzo M, Stancanello J, El Naqa I. Beyond imaging: The promise of radiomics. Phys Med. 2017;38:122-139.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 244]  [Cited by in RCA: 346]  [Article Influence: 38.4]  [Reference Citation Analysis (4)]
17.  Li W, Zhang L, Tian C, Song H, Fang M, Hu C, Zang Y, Cao Y, Dai S, Wang F, Dong D, Wang R, Tian J. Prognostic value of computed tomography radiomics features in patients with gastric cancer following curative resection. Eur Radiol. 2019;29:3079-3089.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 43]  [Cited by in RCA: 67]  [Article Influence: 8.4]  [Reference Citation Analysis (2)]
18.  Zhang W, Fang M, Dong D, Wang X, Ke X, Zhang L, Hu C, Guo L, Guan X, Zhou J, Shan X, Tian J. Development and validation of a CT-based radiomic nomogram for preoperative prediction of early recurrence in advanced gastric cancer. Radiother Oncol. 2020;145:13-20.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 38]  [Cited by in RCA: 120]  [Article Influence: 20.0]  [Reference Citation Analysis (4)]
19.  Cui Y, Zhang J, Li Z, Wei K, Lei Y, Ren J, Wu L, Shi Z, Meng X, Yang X, Gao X. A CT-based deep learning radiomics nomogram for predicting the response to neoadjuvant chemotherapy in patients with locally advanced gastric cancer: A multicenter cohort study. EClinicalMedicine. 2022;46:101348.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 155]  [Cited by in RCA: 163]  [Article Influence: 40.8]  [Reference Citation Analysis (4)]
20.  Carmeliet P, Jain RK. Angiogenesis in cancer and other diseases. Nature. 2000;407:249-257.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 7237]  [Cited by in RCA: 6451]  [Article Influence: 248.1]  [Reference Citation Analysis (5)]
21.  Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, Ghassemi M, Liu X, Reitsma JB, van Smeden M, Boulesteix AL, Camaradou JC, Celi LA, Denaxas S, Denniston AK, Glocker B, Golub RM, Harvey H, Heinze G, Hoffman MM, Kengne AP, Lam E, Lee N, Loder EW, Maier-Hein L, Mateen BA, McCradden MD, Oakden-Rayner L, Ordish J, Parnell R, Rose S, Singh K, Wynants L, Logullo P. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 1587]  [Cited by in RCA: 2632]  [Article Influence: 1316.0]  [Reference Citation Analysis (17)]
22.  Tonello AS, Capelli G, Bao QR, Marchet A, Farinati F, Pawlik TM, Gregori D, Pucciarelli S, Spolverato G. A nomogram to predict overall survival and disease-free survival after curative-intent gastrectomy for gastric cancer. Updates Surg. 2021;73:1879-1890.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 10]  [Cited by in RCA: 15]  [Article Influence: 3.0]  [Reference Citation Analysis (0)]
23.  Lu J, Xu BB, Zheng HL, Li P, Xie JW, Wang JB, Lin JX, Chen QY, Cao LL, Lin M, Tu RH, Huang ZN, Lin JL, Yao ZH, Zheng CH, Huang CM. Robotic versus laparoscopic distal gastrectomy for resectable gastric cancer: a randomized phase 2 trial. Nat Commun. 2024;15:4668.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 34]  [Cited by in RCA: 44]  [Article Influence: 22.0]  [Reference Citation Analysis (0)]
24.  Zwanenburg A, Vallières M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, Ashrafinia S, Bakas S, Beukinga RJ, Boellaard R, Bogowicz M, Boldrini L, Buvat I, Cook GJR, Davatzikos C, Depeursinge A, Desseroit MC, Dinapoli N, Dinh CV, Echegaray S, El Naqa I, Fedorov AY, Gatta R, Gillies RJ, Goh V, Götz M, Guckenberger M, Ha SM, Hatt M, Isensee F, Lambin P, Leger S, Leijenaar RTH, Lenkowicz J, Lippert F, Losnegård A, Maier-Hein KH, Morin O, Müller H, Napel S, Nioche C, Orlhac F, Pati S, Pfaehler EAG, Rahmim A, Rao AUK, Scherer J, Siddique MM, Sijtsema NM, Socarras Fernandez J, Spezi E, Steenbakkers RJHM, Tanadini-Lang S, Thorwarth D, Troost EGC, Upadhaya T, Valentini V, van Dijk LV, van Griethuysen J, van Velden FHP, Whybra P, Richter C, Löck S. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020;295:328-338.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 3143]  [Cited by in RCA: 3123]  [Article Influence: 520.5]  [Reference Citation Analysis (19)]
25.  Hammarström S. The carcinoembryonic antigen (CEA) family: structures, suggested functions and expression in normal and malignant tissues. Semin Cancer Biol. 1999;9:67-81.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 1026]  [Cited by in RCA: 932]  [Article Influence: 34.5]  [Reference Citation Analysis (5)]
26.  Liu X, Cai H, Wang Y. Prognostic significance of tumour markers in Chinese patients with gastric cancer. ANZ J Surg. 2014;84:448-453.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 25]  [Cited by in RCA: 30]  [Article Influence: 2.5]  [Reference Citation Analysis (0)]
27.  Song XH, Zhang WH, Kai-Liu, Chen XL, Zhao LY, Chen XZ, Kun-Yang, Zhou ZG, Hu JK. Prognostic impact of Borrmann classification on advanced gastric cancer: a retrospective cohort from a single institution in western China. World J Surg Oncol. 2020;18:204.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 29]  [Cited by in RCA: 43]  [Article Influence: 7.2]  [Reference Citation Analysis (4)]
28.  Cai Q, Zhou W, Li J, Ou X, Chen C, Cai S, He W, Xu J, He Y. Association of Preoperative Serum Carcinoembryonic Antigen and Gastric Cancer Recurrence: A Large Cohort Study. J Cancer. 2021;12:397-403.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 2]  [Cited by in RCA: 14]  [Article Influence: 2.8]  [Reference Citation Analysis (0)]
29.  Komatsu S, Ichikawa D, Miyamae M, Kosuga T, Okamoto K, Arita T, Konishi H, Morimura R, Murayama Y, Shiozaki A, Kuriu Y, Ikoma H, Nakanishi M, Fujiwara H, Otsuji E. Positive Lymph Node Ratio as an Indicator of Prognosis and Local Tumor Clearance in N3 Gastric Cancer. J Gastrointest Surg. 2016;20:1565-1571.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 18]  [Cited by in RCA: 31]  [Article Influence: 3.1]  [Reference Citation Analysis (0)]
30.  Xie J, Pang Y, Li X, Wu X. The log odds of negative lymph nodes/T stage: a new prognostic and predictive tool for resected gastric cancer patients. J Cancer Res Clin Oncol. 2021;147:2259-2269.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 8]  [Cited by in RCA: 19]  [Article Influence: 3.8]  [Reference Citation Analysis (0)]
31.  Kang WZ, Zhong YX, Ma FH, Xue LY, Xiong JP, Ma S, Li Y, Xie YB, Quan X, Tian YT. Survival outcomes and prognostic indicators for gastric cancer patients with positive peritoneal wash cytology but no peritoneal metastasis after radical gastrectomy. World J Gastrointest Oncol. 2021;13:24-36.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in CrossRef: 7]  [Cited by in RCA: 7]  [Article Influence: 1.4]  [Reference Citation Analysis (3)]
32.  Meng L, Dong D, Chen X, Fang M, Wang R, Li J, Liu Z, Tian J. 2D and 3D CT Radiomic Features Performance Comparison in Characterization of Gastric Cancer: A Multi-Center Study. IEEE J Biomed Health Inform. 2021;25:755-763.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 106]  [Cited by in RCA: 99]  [Article Influence: 19.8]  [Reference Citation Analysis (0)]
33.  Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Granton P, Zegers CM, Gillies R, Boellard R, Dekker A, Aerts HJ. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48:441-446.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 4971]  [Cited by in RCA: 4496]  [Article Influence: 321.1]  [Reference Citation Analysis (15)]
34.  Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278:563-577.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 6963]  [Cited by in RCA: 6536]  [Article Influence: 653.6]  [Reference Citation Analysis (17)]
35.  Banerjee S, Wang DS, Kim HJ, Sirlin CB, Chan MG, Korn RL, Rutman AM, Siripongsakun S, Lu D, Imanbayev G, Kuo MD. A computed tomography radiogenomic biomarker predicts microvascular invasion and clinical outcomes in hepatocellular carcinoma. Hepatology. 2015;62:792-800.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 301]  [Cited by in RCA: 302]  [Article Influence: 27.5]  [Reference Citation Analysis (7)]
36.  Lee CH, Lee HT, Yeh ST, Tsai CY, Lien HH, Lin CS. A Reappraisal of Lymph Node Dissection for Gastric Adenocarcinoma during Upfront Gastrectomy-An Institutional Report. Ann Ital Chir. 2025;96:228-243.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 2]  [Reference Citation Analysis (0)]
37.  Le Z, Zhang Z, Zhou T, Liu M. Esophagojejunal Anastomosis After Totally Laparoscopic Total Gastrectomy: π-shaped Versus Reverse Puncture Device Anastomosis. Ann Ital Chir. 2025;96:1400-1408.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 2]  [Reference Citation Analysis (0)]
38.  Wang Z, Peng W, Zhang J. The Effect of Early Enteral Nutrition under the ERAS Model on Gastrointestinal and Immune Function Recovery in Patients Undergoing Gastric Tumor Surgery. Ann Ital Chir. 2024;95:1147-1154.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 5]  [Reference Citation Analysis (2)]
39.  Cao D, Yu R, Liu H, Li F, Xie W, Tuo C, He J, Hu C, Luo B, Liu Y, Gong W. Effects of Neoadjuvant Chemotherapy on Early Postoperative Anxiety and Depression in Patients With Gastric Cancer: An Observational Study. Ann Ital Chir. 2025;96:1541-1551.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 1]  [Reference Citation Analysis (0)]
Footnotes

Peer review: Externally peer reviewed.

Peer-review model: Single blind

Specialty type: Oncology

Country of origin: China

Peer-review report’s classification

Scientific quality: Grade A, Grade B, Grade B, Grade B, Grade B, Grade C

Novelty: Grade A, Grade B, Grade B, Grade B, Grade B, Grade C

Creativity or innovation: Grade A, Grade B, Grade B, Grade B, Grade C, Grade C

Scientific significance: Grade A, Grade B, Grade B, Grade B, Grade B, Grade C

P-Reviewer: Liu S, MD, China; Vaithiyam V, Assistant Professor, DM, MD, India; Zhou Z, Doctorate Student, China S-Editor: Wu S L-Editor: A P-Editor: Yang YQ

Write to the Help Desk