Abstract
Background
Accurate preoperative assessment of local invasion remains important in gastric cancer, particularly in distinguishing T3/T4 disease that may alter operative planning and perioperative treatment. This study evaluated the diagnostic performance of transabdominal contrast-enhanced ultrasound (CEUS) with gastric filling for preoperative T staging in surgically treated advanced gastric cancer.
Methods
This retrospective single-center diagnostic accuracy study included consecutive patients who underwent CEUS with gastric filling (CEUS-GF) within 2 weeks before gastrectomy between January 2023 and December 2025 and had postoperative pathologic T2–T4 disease. Two blinded sonographers independently assigned clinical T (cT) stage; discrepant cases were adjudicated for the final imaging classification. Exact agreement, Cohen’s κ, stage-specific performance, binary discrimination, receiver operating characteristic (ROC) analysis, and subgroup accuracy were assessed against postoperative pathology as the reference standard.
Results
In this cohort of 180 patients, exact agreement between final CEUS stage and pathology was 84.4%, with substantial agreement after grouping T4a and T4b as T4 (κ=0.781); initial inter-reader agreement was κ=0.822. For identifying pathologic T4 disease, sensitivity was 95.8%, specificity was 94.4%, positive predictive value was 92.0%, negative predictive value was 97.1%, and accuracy was 95.0%. Stage-specific accuracy was 90.0% for T2, 87.2% for T3, and 95.0% for grouped T4 disease. The area under the ROC curve (AUC) was 0.876 for discrimination of pathologic T stage (pT) 3-or-higher disease from pT2 and 0.966 for discrimination of pT4 from non-T4 disease. Exact T4a/T4b subclassification accuracy was 91.7%. In descriptive subgroup analyses, accuracy was numerically highest for gastric body lesions (91.8%).
Conclusions
In this surgically treated single-center T2–T4 cohort, CEUS-GF showed good agreement with pathology and high discrimination for pathologic T4 disease. CEUS-GF may be a useful adjunct for assessing suspected serosal or extragastric extension before surgery; prospective multicenter paired-comparison studies are needed to define its incremental value beyond computed tomography (CT)/endoscopic ultrasound (EUS).
Keywords: Contrast-enhanced ultrasound (CEUS), gastric cancer, T staging, gastric filling, Borrmann classification
Highlight box.
Key findings
• Contrast-enhanced transabdominal ultrasound with gastric filling (CEUS-GF) achieved 84.4% exact agreement with postoperative pathology for T2–T4 gastric cancer staging (κ=0.781 after grouping T4a/b). For detecting pT4 disease, sensitivity was 95.8%, specificity 94.4%, and the area under the receiver operating characteristic curve (AUC) was 0.966. Stage-specific accuracy was 90.0% (T2), 87.2% (T3), and 95.0% (T4 grouped). In descriptive subgroup analyses, accuracy was numerically highest for body lesions (91.8%).
What is known and what is new?
• Preoperative differentiation of T3 vs. T4 gastric cancer is critical for surgical planning and neoadjuvant therapy decisions. Endoscopic ultrasound (EUS) and computed tomography (CT) have limited accuracy for serosal invasion, and transabdominal ultrasound is often considered operator-dependent and insufficiently reliable.
• In this large single-center surgical cohort, CEUS-GF showed high agreement with pathology for T2–T4 staging and excellent inter-reader agreement, supporting its potential as a reproducible, non- invasive adjunct; external validation is needed to define its clinical role. It demonstrates that CEUS-GF can identify T4 disease with high sensitivity and specificity, and that subclassification into T4a and T4b remains feasible in most cases.
What is the implication, and what should change now?
• CEUS-GF could serve as an accessible, radiation-free adjunct to preoperative local staging, particularly for assessing suspected serosal or extragastric extension when conventional imaging findings are indeterminate. Its high agreement with pathology and inter-reader reproducibility support further clinical evaluation. Prospective multicenter paired-comparison studies are needed to define its incremental value beyond CT/EUS and its role in treatment planning.
Introduction
Gastric cancer remains a major cause of cancer morbidity and mortality worldwide. The Global Cancer Observatory (GLOBOCAN) 2022 estimated that stomach cancer accounted for 4.9% of incident cancers and 6.8% of cancer deaths globally, underscoring a persistent burden despite regional declines in incidence (1). Contemporary treatment is explicitly stage-dependent. Patients with disease confined to the stomach may be managed with surgery or endoscopic resection in carefully selected settings, whereas locally advanced tumors usually require gastrectomy integrated with perioperative systemic therapy, and metastatic disease is approached with palliative systemic treatment guided by biomarkers and performance status (2,3). Because these treatment pathways diverge according to the depth of mural invasion and the likelihood of serosal or adjacent organ involvement, reliable preoperative T staging remains central to clinical decision-making in gastric cancer (2,3).
In routine practice, many patients are diagnosed after symptoms appear, when the tumor has already extended beyond the mucosa and decisions about resectability, surgical extent, and perioperative treatment become more complex (2). Distinguishing T2 from T3 disease helps estimate the risk of subserosal extension, while identifying T4a or T4b disease has immediate implications for the assessment of serosal breach, possible invasion of neighboring structures, and the need for broader operative planning (3). The problem is especially relevant in gastric cancer with deep mural invasion or suspected serosal involvement, where small errors in estimating local invasion may lead to understaging, unnecessary laparotomy, or overestimation of organ invasion that is not confirmed pathologically. An imaging method used in this setting should therefore combine reproducible mural visualization with adequate sensitivity for serosal and extramural extension.
Contrast-enhanced computed tomography (CT) remains the main cross-sectional technique for initial staging because it surveys the entire abdomen and evaluates distant disease in the same examination (3). Its weakness lies in limited discrimination of the fine gastric wall layers, particularly when the stomach is poorly distended or when inflammatory change obscures the outer contour (4). Endoscopic ultrasound (EUS) offers higher spatial resolution for locoregional assessment and is often more informative for early mural invasion, yet performance varies with operator expertise, tumor location, luminal stenosis, and bulky infiltrative lesions (4-6). In a 2023 multicenter study, linear EUS outperformed contrast-enhanced CT for T and N staging, but accuracy still ranged from 60% to 80% across T2–T4 categories and the technique remained dependent on endoscopic access (5). A 2024 review similarly emphasized that EUS is strongest for locoregional staging but is less useful for distant metastatic assessment (6). Conventional transabdominal ultrasound is attractive because it is noninvasive, repeatable, and dynamically performed, although gas within the stomach often degrades image quality and masks mural interfaces.
Gastric filling addresses one of the main technical limitations of transabdominal ultrasound by distending the lumen, displacing intragastric gas, and improving visualization of the mural stratification. When intravenous microbubble contrast is added, the examination can depict enhancement boundaries, interruption of the serosal line, perigastric fat infiltration, and suspected direct extension beyond the gastric wall. Recent evidence supports the biological plausibility of this approach. A 2022 meta-analysis of preoperative studies reported pooled sensitivities of 0.81 for T2, 0.88 for T3, and 0.81 for T4 disease, with corresponding specificities of 0.96, 0.85, and 0.96, suggesting that contrast-enhanced ultrasound (CEUS) with gastric filling can achieve good discrimination across clinically relevant stages (7). In a prospective comparison published in 2023, oral contrast-enhanced transabdominal ultrasound achieved overall T-stage accuracy comparable to enhanced CT (72.2% vs. 75.9%), performed better in T1-T2 disease, and remained clinically informative in routine preoperative assessment (8). A 2024 clinical study of oral CEUS reported mean T-stage accuracy of 75.5%, with accuracy approaching 88% for T4a and T4b lesions (9). A 2025 meta-analysis comparing enhanced CT with CEUS-based gastric ultrasound also suggested higher pooled sensitivity for CEUS, particularly at the extremes of T staging, although the available literature remained heterogeneous (10).
Even so, the published evidence is still dominated by mixed-stage cohorts, relatively small single-center series, and studies designed around early plus advanced disease together rather than the subset in which local invasion is the dominant clinical question (8-10). Data focused on consecutive patients with postoperative pathologic T2–T4 gastric cancer are more limited, and subgroup effects related to tumor site, Borrmann morphology, or contrast-agent selection have not been sufficiently characterized in routine practice. The present study was designed to evaluate the diagnostic performance of contrast-enhanced transabdominal ultrasound with gastric filling for preoperative T staging in a consecutive cohort of patients with pathologic T2–T4 gastric cancer, using postoperative pathology as the reference standard. Emphasis was placed on exact T-stage agreement, discrimination of pathologic T4 disease, inter-reader agreement, and prespecified subgroup analyses that may influence clinical applicability. We present this article in accordance with the STARD reporting checklist (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0800/rc).
Methods
Study design and patients
This single-center retrospective diagnostic accuracy study was performed at Shengli Oilfield Central Hospital. Consecutive patients evaluated between January 2023 and December 2025 were screened through the hospital information system, picture archiving and communication system, endoscopy database, and pathology registry. The study aimed to assess the performance of contrast-enhanced transabdominal ultrasound with gastric filling for preoperative T staging of advanced gastric cancer in routine care. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The institutional ethics committee of Shengli Oilfield Central Hospital approved the study protocol (No. YXLL202513202) and waived the requirement for written informed consent because the analysis used de-identified retrospective data and did not alter patient management.
Eligible patients met all of the following criteria: age 18–80 years; gastric cancer confirmed on endoscopic biopsy; CEUS with gastric filling performed within 2 weeks before surgery; curative or palliative gastrectomy with postoperative pathological T staging available; and complete imaging, pathology, and clinical records. Patients were excluded if they had pathologic T1 disease, received neoadjuvant chemotherapy, radiotherapy, targeted therapy, or immunotherapy before ultrasound evaluation, had a history of another active malignancy, had image quality insufficient for confident mural-layer or serosal assessment, or lacked key clinical or pathological data. After application of these criteria, 180 patients with postoperative pathologic T2–T4 gastric cancer comprised the final study cohort.
Data collection
Two investigators independently extracted demographic, clinical, imaging, operative, and pathological data using a predefined case-report form and resolved discrepancies by source-document review. Variables collected for baseline description included age, sex, body mass index, presenting symptoms, tumor site, tumor size, histologic differentiation, Borrmann type, contrast agent used for CEUS, enhancement pattern, serosal appearance, perigastric fat-plane findings, and pathological T stage. Tumor site was grouped as fundus/cardia, body, antrum, or whole stomach. For subgroup analyses, Borrmann type I (polypoid/fungating), type II (ulcerative), type III (ulceroinfiltrative), and type IV (diffusely infiltrative) were assigned according to the final gross-pathology report; when the original pathology record used descriptive macroscopic terminology rather than Roman numerals, the corresponding Borrmann category was recorded accordingly.
CEUS with gastric filling protocol
All examinations were performed after at least 8 hours of fasting by gastrointestinal ultrasound physicians using high-end ultrasound systems (Mindray, China) equipped with contrast-specific imaging software. A 3.5–5.0 MHz convex transducer was used for the initial abdominal survey, and a 7.5–12.0 MHz linear transducer was added when superficial segments of the anterior wall or serosal contour required closer evaluation. After baseline gray-scale and color Doppler scanning, patients ingested 500–800 mL of Tianxia (Huzhou East Asia Medical Products, Huzhou, China), an oral ultrasound contrast agent, to achieve satisfactory gastric distention. The stomach was scanned in multiple positions, including supine and left lateral decubitus, to optimize visualization of the lesion, mural stratification, and perigastric interface.
Intravenous CEUS was then performed with a low mechanical index setting (<0.12). According to routine clinical availability, either SonoVue (sulfur hexafluoride microbubbles, 59 mg; Bracco Suisse SA, Plan-Les-Ouates, Switzerland) or LidaXing (perfluoropropane microbubbles, 0.5 g; Xiamen Lizhuo Pharmaceutical, Xiamen, China) was administered through a peripheral vein, followed by a saline flush. Continuous cine acquisition began immediately after injection and extended for at least 3 minutes. Enhancement findings were reviewed during the early arterial phase (10–30 s), portal-venous phase (30–120 s), and late phase (>120 s). Recorded sonographic variables included mural thickness and stratification, enhancement pattern, continuity of the outer hyperechoic serosal line, irregularity of the external gastric contour, perigastric fat-plane change, and suspected invasion of adjacent organs.
Clinical T stage was assigned by integrating gray-scale, gastric-filling, and contrast-enhanced features, with postoperative pathology as the reference. Lesions were categorized as cT2 when invasion involved the muscularis propria with preservation of a smooth outer hyperechoic serosal layer. cT3 was assigned when mural stratification was largely lost and the lesion extended to the subserosa, but the outer serosal line remained continuous without definite extramural extension. cT4a was diagnosed when the serosal line was interrupted, blurred, or focally obliterated, especially when accompanied by nodular or spiculated extension into perigastric fat. cT4b was diagnosed when the fat plane between the tumor and an adjacent organ disappeared and the lesion showed fixed broad-based contact or direct continuity with that organ. These criteria were based on the American Joint Committee on Cancer (AJCC) 8th edition pathologic framework and on previously described gastric ultrasound staging features (8,9,11).
Image interpretation and reference standard
All stored ultrasound examinations were reviewed independently by two attending sonographers, each with more than 5 years of experience in gastrointestinal CEUS, who were blinded to the postoperative pathological findings and to each other’s assessments. During the retrospective image review, readers had no access to clinical or endoscopic information. Each reader assigned a clinical T stage and recorded the key imaging signs used for classification. When the two initial interpretations differed, the images were re-reviewed in a joint session to identify the basis of disagreement. Cases that still lacked consensus were adjudicated by a senior gastrointestinal ultrasound physician with more than 10 years of CEUS experience who had not participated in the initial reads. Initial inter-reader agreement was quantified to reflect the reproducibility of the imaging assessment, and the adjudicated result was used as the final imaging classification for the diagnostic analyses.
The reference standard was postoperative pathological T staging of the resected specimen according to the AJCC Cancer Staging Manual, 8th edition (11). T2 was defined as invasion of the muscularis propria, T3 as penetration into the subserosal connective tissue without invasion of the visceral peritoneum or adjacent structures, T4a as invasion of the serosa, and T4b as direct invasion of adjacent organs or structures. Pathological stage and gross morphology were reviewed by two gastrointestinal pathologists within routine practice, with discrepancies resolved by joint review before final sign-out.
Study outcomes and statistical analysis
The prespecified primary outcomes were exact agreement between the final imaging T stage and pathological T stage, Cohen’s κ for imaging-pathology agreement, and the diagnostic performance of CEUS with gastric filling for identifying pathologic T4 disease (12). Because κ corrects for agreement expected by chance, it is widely used as a reliability measure for categorical diagnostic ratings; values of 0.41–0.60, 0.61–0.80, and 0.81–1.00 were interpreted as moderate, substantial, and near-perfect agreement, respectively, using the Landis-Koch framework (13). Secondary outcomes included stage-specific sensitivity, specificity, positive predictive value, negative predictive value, and accuracy for T2, T3, and T4 categories; binary discrimination of T3-or-higher versus T2 disease; accuracy of T4a/T4b subclassification; rates of overstaging (cT > pT) and understaging (cT < pT); and subgroup exact-staging accuracy according to tumor site, Borrmann type, and intravenous contrast agent.
Statistical analysis
Statistical analysis was performed using SPSS version 26.0 and R software. Continuous variables are presented as mean ± standard deviation when approximately normally distributed and as median (interquartile range) otherwise; categorical variables are presented as counts and percentages. Exact agreement was calculated as the proportion of patients in whom imaging and pathology assigned the same T category. Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were reported with 95% confidence intervals (CIs) calculated by the Wilson score method, which provides stable interval estimation for binomial proportions in moderate-sized samples (14). Two-sided 95% CIs for Cohen’s κ were estimated by nonparametric patient-level bootstrap resampling (10,000 replicates; percentile method). For T4 versus non-T4 discrimination, positive and negative likelihood ratios and the diagnostic odds ratio were calculated from the 2×2 table, with 95% CIs calculated on the log scale. Subgroup analyses were descriptive and report exact-staging accuracy with Wilson 95% CIs. Missing values were not imputed; only patients with complete imaging and pathological reference data were included in the final analysis.
Results
Baseline clinical and pathologic characteristics
As shown in Figure 1, of 286 patients who underwent CEUS with gastric filling for gastric cancer between January 2023 and December 2025, 106 were excluded for the following reasons: 21 did not undergo gastrectomy, 8 lacked postoperative pathologic T staging, 21 received neoadjuvant therapy before ultrasound, 18 had pathologic T1 disease, 17 had incomplete data, 14 had insufficient image quality for T staging, and 7 had another active malignancy. The final cohort comprised 180 patients with pathologically confirmed advanced gastric cancer (T2–T4).
Figure 1.
Study flow diagram. Of 286 patients who underwent contrast-enhanced ultrasound with gastric filling for gastric cancer between January 2023 and December 2025, 180 patients with pathologically confirmed advanced gastric cancer (T2–T4) were included in the final analysis. CEUS, contrast-enhanced ultrasound; T, tumor.
As summarized in Table 1, the mean age of the cohort was 60.9 years, the mean body mass index was 22.5 kg/m2, and the mean tumor size was 5.6 cm; 127 patients (70.6%) were men. Tumors were located in the antrum in 62 patients (34.4%), in the gastric body in 61 (33.9%), in the fundus/cardia in 43 (23.9%), and in the whole stomach in 14 (7.8%). Histologic differentiation was poorly differentiated in 19 patients (10.6%), moderately differentiated in 90 (50.0%), and well differentiated in 71 (39.4%). Borrmann type III was the most common gross type, occurring in 75 patients (41.7%), followed by type II in 52 (28.9%), type IV in 36 (20.0%), and type I in 17 (9.4%). SonoVue and LidaXing were used in 131 (72.8%) and 49 examinations (27.2%), respectively.
Table 1. Baseline clinical, imaging, and pathologic characteristics of the study cohort.
| Characteristic | Overall cohort (N=180) |
|---|---|
| Demographic characteristics | |
| Age, years | 60.90±8.60 |
| Body mass index, kg/m2 | 22.50±2.20 |
| Sex | |
| Male | 127 (70.56) |
| Female | 53 (29.44) |
| Tumor-related characteristics | |
| Tumor site | |
| Antrum | 62 (34.44) |
| Body | 61 (33.89) |
| Fundus/cardia | 43 (23.89) |
| Whole stomach | 14 (7.78) |
| Tumor size, cm | 5.60±1.50 |
| Histologic differentiation | |
| Poorly differentiated | 19 (10.56) |
| Moderately differentiated | 90 (50.00) |
| Well differentiated | 71 (39.44) |
| Borrmann type | |
| Type I (polypoid/fungating) | 17 (9.44) |
| Type II (ulcerative) | 52 (28.89) |
| Type III (ulceroinfiltrative) | 75 (41.67) |
| Type IV (diffusely infiltrative) | 36 (20.00) |
| Contrast agent | |
| SonoVue | 131 (72.78) |
| LidaXing | 49 (27.22) |
| Clinical symptoms | |
| Abdominal pain | 115 (63.89) |
| Acid reflux | 58 (32.22) |
| Weight loss | 90 (50.00) |
| Gastrointestinal bleeding | 51 (28.33) |
| CEUS findings | |
| Enhancement pattern | |
| Hypoenhancement | 35 (19.44) |
| Homogeneous enhancement | 82 (45.56) |
| Marked enhancement | 63 (35.00) |
| Serosal layer appearance | |
| Intact | 66 (36.67) |
| Interrupted or blurred | 63 (35.00) |
| Signs of extramural invasion | 51 (28.33) |
| Perigastric fat-plane appearance | |
| Clear | 52 (28.89) |
| Mildly obscured | 82 (45.56) |
| Infiltration or nodularity | 46 (25.56) |
| Reference standard | |
| Pathologic T stage | |
| T2 | 34 (18.89) |
| T3 | 74 (41.11) |
| T4a | 48 (26.67) |
| T4b | 24 (13.33) |
Data are presented as mean ± SD or number (%). Percentages were calculated with N=180 as the denominator. CEUS, contrast-enhanced ultrasound; SD, standard deviation; T, tumor.
Abdominal pain, acid reflux, weight loss, and gastrointestinal bleeding were reported in 115 (63.9%), 58 (32.2%), 90 (50.0%), and 51 patients (28.3%), respectively. On CEUS, homogeneous enhancement was the most common pattern (82/180, 45.6%), followed by marked enhancement (63/180, 35.0%) and hypoenhancement (35/180, 19.4%). The serosal layer was intact in 66 patients (36.7%), interrupted or blurred in 63 (35.0%), and suggestive of extramural invasion in 51 (28.3%). The perigastric fat plane was clear in 52 patients (28.9%), mildly obscured in 82 (45.6%), and showed infiltration or nodularity in 46 (25.6%). According to postoperative pathology, 34 tumors (18.9%) were classified as T2, 74 (41.1%) as T3, 48 (26.7%) as T4a, and 24 (13.3%) as T4b.
Diagnostic performance for identifying pathologic T4 disease
At the patient level, final CEUS-based cT staging exactly matched postoperative pathologic staging in 152 of 180 patients, corresponding to an overall exact accuracy of 84.4%. Agreement between final CEUS staging and pathology at the grouped T2/T3/T4 level was κ=0.781 (95% CI, 0.698–0.855), and initial inter-reader agreement was κ=0.822 (95% CI, 0.742–0.893). Table 2 summarizes the diagnostic performance of final CEUS-based cT staging for identifying pathologic T4 disease. Sensitivity was 95.8% (95% CI, 88.5–98.6%), specificity was 94.4% (95% CI, 88.4–97.4%), positive predictive value was 92.0% (95% CI, 83.6–96.3%), and negative predictive value was 97.1% (95% CI, 91.9–99.0%). The overall accuracy for discriminating T4 from non-T4 disease was 95.0% (95% CI, 90.8–97.3%). The positive likelihood ratio was 17.25 (95% CI, 7.92–37.60), the negative likelihood ratio was 0.044 (95% CI, 0.015–0.134), and the diagnostic odds ratio was 391.0 (95% CI, 94.6–1,616.3).
Table 2. Diagnostic performance of final CEUS-based cT staging for identifying pathologic T4 disease.
| Metric | Numerator/denominator | Estimate, % | 95% CI, % |
|---|---|---|---|
| Sensitivity | 69/72 | 95.83 | 88.45–98.57 |
| Specificity | 102/108 | 94.44 | 88.41–97.43 |
| Positive predictive value | 69/75 | 92.00 | 83.63–96.28 |
| Negative predictive value | 102/105 | 97.14 | 91.93–99.02 |
| Accuracy | 171/180 | 95.00 | 90.77–97.35 |
| Positive likelihood ratio (LR+) | – | 17.25 | 7.92–37.60 |
| Negative likelihood ratio (LR−) | – | 0.044 | 0.015–0.134 |
| Diagnostic odds ratio | – | 391.0 | 94.6–1,616.3 |
T4 denotes combined T4a and T4b disease. Final cT stage was compared against pathologic T stage as the reference standard. All 95% confidence intervals were calculated by the Wilson method. CEUS, contrast-enhanced ultrasound; CI, confidence interval; cT, clinical T stage; T, tumor.
Overall concordance between final CEUS staging and pathologic staging
Figure 2 shows the grouped confusion matrix for final CEUS-based cT staging versus pathologic pT staging after collapsing T4a and T4b into a single T4 category. Exact grouped agreement was observed in 24 of 34 pathologic T2 tumors, 62 of 74 pathologic T3 tumors, and 69 of 72 pathologic T4 tumors. Misclassification of pathologic T2 disease most often involved assignment to T3 (8 cases), with 2 additional T2 tumors overstaged as T4. Among pathologic T3 tumors, 8 were understaged as T2 and 4 were overstaged as T4. Among grouped pathologic T4 tumors, 3 cases were classified as T3 and none were classified as T2.
Figure 2.
Patient-level confusion matrix comparing final CEUS-based cT stage with pathologic pT stage. Rows indicate pathologic stage and columns indicate final CEUS-based cT stage. Cell values denote the numbers of patients in each classification cell. T4 denotes combined T4a and T4b disease. CEUS, contrast-enhanced ultrasound; cT, clinical T stage; pT, pathologic T stage; T, tumor.
In the full four-category comparison, exact subclassification was achieved in 66 of 72 pathologic T4 tumors (91.7%). The complete T2/T3/T4a/T4b confusion matrix is provided in Table S1. Specifically, 44 of 48 pathologic T4a tumors were classified as T4a, 3 were classified as T3, and 1 was classified as T4b, whereas 22 of 24 pathologic T4b tumors were classified as T4b and 2 were classified as T4a.
Stage-specific diagnostic performance
As shown in Table 3, stage-specific performance varied across T categories. For T2 disease, sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were 70.6%, 94.5%, 75.0%, 93.2%, and 90.0%, respectively. For T3 disease, the corresponding values were 83.8%, 89.6%, 84.9%, 88.8%, and 87.2%. For grouped T4 disease, sensitivity was 95.8%, specificity was 94.4%, positive predictive value was 92.0%, negative predictive value was 97.1%, and accuracy was 95.0%.
Table 3. Stage-specific diagnostic performance of final CEUS-based cT staging.
| Stage | Metric | Numerator/denominator | Estimate, % | 95% CI, % |
|---|---|---|---|---|
| T2 | Sensitivity | 24/34 | 70.59 | 53.83–83.17 |
| Specificity | 138/146 | 94.52 | 89.56–97.20 | |
| Positive predictive value | 24/32 | 75.00 | 57.89–86.75 | |
| Negative predictive value | 138/148 | 93.24 | 88.01–96.29 | |
| Accuracy | 162/180 | 90.00 | 84.75–93.58 | |
| T3 | Sensitivity | 62/74 | 83.78 | 73.76–90.47 |
| Specificity | 95/106 | 89.62 | 82.37–94.11 | |
| Positive predictive value | 62/73 | 84.93 | 75.00–91.37 | |
| Negative predictive value | 95/107 | 88.79 | 81.41–93.47 | |
| Accuracy | 157/180 | 87.22 | 81.56–91.33 | |
| T4 | Sensitivity | 69/72 | 95.83 | 88.45–98.57 |
| Specificity | 102/108 | 94.44 | 88.41–97.43 | |
| Positive predictive value | 69/75 | 92.00 | 83.63–96.28 | |
| Negative predictive value | 102/105 | 97.14 | 91.93–99.02 | |
| Accuracy | 171/180 | 95.00 | 90.77–97.35 |
T4 denotes combined T4a and T4b disease. Final cT stage was compared against pathologic T stage as the reference standard. All 95% confidence intervals were calculated by the Wilson method. CEUS, contrast-enhanced ultrasound; CI, confidence interval; cT, clinical T stage; T, tumor.
Visual summary of diagnostic performance
Figure 3 provides a visual summary of the principal diagnostic estimates. All T4-versus-non-T4 metrics exceeded 90.0%, and the stage-specific accuracy estimates were 90.0% for T2, 87.2% for T3, and 95.0% for T4. Receiver operating characteristic (ROC) analysis using the ordinal final CEUS stage as the test variable yielded an area under the ROC curve (AUC) of 0.876 for discrimination of pT3-or-higher disease from pT2 and 0.966 for discrimination of pT4 from non-T4 disease (Figure 4), again indicating better separation for T4 disease.
Figure 3.
Forest plot of diagnostic performance estimates with 95% confidence intervals. The plot shows sensitivity, specificity, positive predictive value, negative predictive value, and accuracy for discrimination of pathologic T4 versus non-T4 disease, together with stage-specific accuracy estimates for T2, T3, and T4. Squares indicate point estimates and horizontal lines indicate 95% confidence intervals. NPV, negative predictive value; PPV, positive predictive value; T, tumor.
Figure 4.
Receiver operating characteristic curves for binary discrimination using the ordinal final CEUS-based cT stage. The AUC was 0.876 for pT3-or-higher versus pT2 disease and 0.966 for pT4 versus non-T4 disease. Annotated points indicate the clinically relevant thresholds cT ≥ T3 (disease beyond T2) and cT ≥ T4 (suspected serosal or adjacent-organ involvement). These curves reflect ordinal discrimination, not calibration or clinical net benefit. AUC, area under the receiver operating characteristic curve; CEUS, contrast-enhanced ultrasound; cT, clinical T stage; pT, pathologic T stage; T, tumor.
Clinically relevant binary staging analyses
Table 4 presents the binary staging analyses. For discrimination of T3-or-higher disease from T2 disease, sensitivity was 94.5%, specificity was 70.6%, positive predictive value was 93.2%, negative predictive value was 75.0%, and overall accuracy was 90.0%. For discrimination of T4 from non-T4 disease, sensitivity was 95.8%, specificity was 94.4%, positive predictive value was 92.0%, negative predictive value was 97.1%, and overall accuracy was 95.0%. Additional staging-deviation metrics showed a T4a/T4b subclassification accuracy of 91.7%, an overstaging rate of 8.3%, and an understaging rate of 7.2%. Taken together, these binary estimates indicate stronger separation for T4 versus non-T4 disease than for T3-or-higher versus T2 disease.
Table 4. Diagnostic performance in clinically relevant binary staging analyses.
| Binary analysis | Metric | Numerator/denominator | Estimate, % | 95% CI, % |
|---|---|---|---|---|
| ≥T3 vs. T2 | Sensitivity | 138/146 | 94.52 | 89.56–97.20 |
| Specificity | 24/34 | 70.59 | 53.83–83.17 | |
| Positive predictive value | 138/148 | 93.24 | 88.01–96.29 | |
| Negative predictive value | 24/32 | 75.00 | 57.89–86.75 | |
| Accuracy | 162/180 | 90.00 | 84.75–93.58 | |
| T4 vs. non-T4 | Sensitivity | 69/72 | 95.83 | 88.45–98.57 |
| Specificity | 102/108 | 94.44 | 88.41–97.43 | |
| Positive predictive value | 69/75 | 92.00 | 83.63–96.28 | |
| Negative predictive value | 102/105 | 97.14 | 91.93–99.02 | |
| Accuracy | 171/180 | 95.00 | 90.77–97.35 |
All 95% confidence intervals were calculated by the Wilson method. The reported accuracy for T4a/T4b subclassification was 91.67%, with overall overstaging and understaging rates of 8.33% and 7.22%, respectively. CI, confidence interval; T, tumor.
Subgroup analyses
Table 5 and Figure 5 summarize the descriptive subgroup analyses. Exact-staging accuracy was numerically highest for lesions in the gastric body (91.8%), followed by fundus/cardia lesions (86.0%), whole-stomach lesions (78.6%), and antral lesions (77.4%). Accuracy estimates for Borrmann type and contrast-agent subgroups are also presented with 95% CIs. Tumor-location differences were statistically significant (chi-square =5.296, P=0.02). According to Borrmann classification, accuracy was 82.4% for type I (polypoid/fungating) tumors, 76.9% for type II (ulcerative) tumors, 89.3% for type III (ulceroinfiltrative) tumors, and 86.1% for type IV (diffusely infiltrative) tumors; this difference did not reach statistical significance (chi-square =3.737, P=0.053). According to contrast agent, exact-staging accuracy was 85.5% in examinations performed with SonoVue and 81.6% in those performed with LidaXing, without a significant between-group difference (chi-square =0.405, P=0.52).
Table 5. Diagnostic accuracy according to prespecified subgroups.
| Subgroup factor | Category | No. of patients | Correctly staged, n/total | Accuracy, % | 95% CI, % |
|---|---|---|---|---|---|
| Tumor site | Antrum | 62 | 48/62 | 77.42 | 65.56–85.97 |
| Body | 61 | 56/61 | 91.80 | 82.19–96.35 | |
| Fundus/cardia | 43 | 37/43 | 86.05 | 72.66–93.41 | |
| Whole stomach | 14 | 11/14 | 78.57 | 52.36–92.43 | |
| Borrmann type | Type I (polypoid/fungating) | 17 | 14/17 | 82.35 | 58.96–93.82 |
| Type II (ulcerative) | 52 | 40/52 | 76.92 | 63.87–86.30 | |
| Type III (ulceroinfiltrative) | 75 | 67/75 | 89.33 | 80.30–94.49 | |
| Type IV (diffusely infiltrative) | 36 | 31/36 | 86.11 | 71.33–93.95 | |
| Contrast agent | SonoVue | 131 | 112/131 | 85.50 | 78.45–90.52 |
| LidaXing | 49 | 40/49 | 81.63 | 68.61–89.99 |
Accuracy was defined as exact agreement between the final cT stage and the pathologic T stage within each subgroup. Subgroup analyses were descriptive. All 95% confidence intervals were calculated by the Wilson method. CI, confidence interval; cT, clinical T stage; T, tumor.
Figure 5.
Subgroup accuracy according to tumor site, Borrmann classification, and contrast agent. Horizontal bars represent exact-staging accuracy for each subgroup category. Values are shown as percentages and correspond to the patient counts summarized in Table 5.
Discussion
In this retrospective cohort, we found that CEUS with gastric filling achieved good patient-level agreement with pathology for preoperative T staging of locally advanced gastric cancer, with an exact accuracy of 84.4% and κ=0.781. The strongest performance was observed for T4 disease, with sensitivity of 95.8%, specificity of 94.4%, accuracy of 95.0%, and an AUC of 0.966. By comparison, discrimination of pT3-or-higher disease from pT2 was weaker, although still acceptable, with an AUC of 0.876. This pattern is clinically relevant because local T stage affects operative planning, resectability assessment, and perioperative treatment pathways in gastric cancer (3,15). We deliberately excluded pathologic T1 disease because the present study was designed to address a different clinical problem, namely preoperative assessment of deeper mural invasion in patients proceeding to gastrectomy, in whom discrimination of T3/T4 disease and recognition of serosal or adjacent-organ involvement are more relevant than early-stage layer-by-layer staging.
Our findings are broadly consistent with published work on oral or double contrast-enhanced gastric ultrasonography. Zhang et al. reported in a meta-analysis that double contrast-enhanced gastric ultrasonography (DCEUS) had pooled sensitivities and specificities of 0.88 and 0.85 for T3 disease and 0.81 and 0.96 for T4 disease (7). A 2024 meta-analysis by Zhong et al. similarly supported favorable pooled diagnostic performance for contrast-enhanced gastric ultrasonography in gastric T staging (16). Wu et al. reported a prospective comparison of oral contrast-enhanced transabdominal ultrasound with contrast-enhanced CT in preoperative T staging (8), and Liang et al. also reported clinical utility for preoperative T-stage assessment (9). A more recent meta-analysis comparing DCEUS with enhanced CT reported pooled diagnostic estimates for both modalities in gastric T staging (10). These external studies do not establish incremental value over CT or EUS in our cohort, although direct cross-study comparisons should be made cautiously because stage spectrum, equipment, and reader experience differ across studies (16).
From an imaging perspective, gastric filling improves luminal distension and reduces interference from gas, which helps display gastric wall stratification and lesion extent. Intravenous contrast then adds real-time information on mural perfusion, serosal interruption, and perigastric fat-plane change. These features are especially useful when the diagnostic task is to recognize transmural or extragastric extension rather than to separate smaller intramural differences. This stage-dependent behavior is consistent with the broader literature on gastric T staging, where the T2–T3 and T3–T4 interfaces remain the main sources of error across EUS, CT, and ultrasound-based approaches (17,18). A recent retrospective ultrasound study that focused on T3–T4 disease also emphasized the importance of mural disruption and perigastric findings in identifying locally advanced tumors (19).
The lower sensitivity for T2 disease and the modest reduction in T3 accuracy are clinically understandable. Overstaging may occur when peritumoral inflammation, fibrosis, or edema obscure the outer wall and simulate serosal involvement, whereas understaging can arise when microserosal invasion is focal or when the insonation plane does not capture the deepest area of extension. Our confusion matrix reflected both mechanisms: most T2 errors shifted upward to T3, while a small number of pathologic T4 tumors were classified as T3. The 91.7% exact accuracy for T4a/T4b subclassification is encouraging, but it also indicates that separation of serosal penetration from definite invasion of adjacent structures remains imperfect on transabdominal imaging alone. This limitation matters because T4b disease may alter resectability assessment and operative strategy (3,15,18).
Although descriptive, the subgroup analysis adds nuance and should not be overinterpreted. Accuracy was highest in gastric body lesions and was lower in antral and whole-stomach disease, suggesting that local acoustic conditions and lesion extent can materially affect staging performance. Borrmann type III (ulceroinfiltrative) tumors showed the highest numerical accuracy. Borrmann morphology is clinically meaningful in advanced gastric cancer (20), but the present descriptive subgroup analysis was not designed to identify independent determinants of staging accuracy. Contrast-agent selection was pragmatic rather than randomized; therefore, the observed subgroup estimates should be interpreted as exploratory.
Several limitations should temper interpretation. This was a retrospective single-center study, and only patients with surgery and complete imaging-pathology data were included, which may have introduced selection bias. This selected, surgically treated cohort may have yielded more favorable sensitivity, specificity, and overall accuracy estimates than would be observed in an unselected clinical population. Because T1 disease was excluded and all participants proceeded to gastrectomy, these findings do not estimate performance for early gastric cancer, patients receiving neoadjuvant therapy, or patients managed without surgery. The pathological T-stage distribution in this cohort also influences positive and negative predictive values. The sample size was reasonable overall but limited for some subgroups, especially whole-stomach lesions and T4b disease. Detailed individual training histories and cumulative CEUS case volumes were not prospectively or systematically recorded in this retrospective study and therefore could not be reliably reconstructed, which may limit assessment of reproducibility across centers. Moreover, the use of a single equipment manufacturer, only two contrast agents, and highly experienced readers may further limit generalizability to centers with different platforms, contrast protocols, or levels of expertise. Only 28 patients had an incorrect exact T-stage classification; therefore, a multivariable model including tumor site, Borrmann type, tumor size, and differentiation would be overfitted. We also did not perform a head-to-head comparison with CT or EUS in the same cohort; therefore, incremental value over established staging modalities cannot be determined. In addition, the ROC analysis was based on ordinal final cT categories rather than on a continuous prediction model; therefore, the AUC values summarize stage-order discrimination rather than calibrated risk estimation. Because the index test provided ordinal T-stage categories rather than individual predicted probabilities with prespecified clinical action thresholds, calibration and decision-curve analyses were not undertaken. Prospective multicenter studies with standardized acquisition and blinded image review are needed to clarify how CEUS with gastric filling should be integrated with CT and EUS in routine staging pathways(15).
Conclusions
In this surgically treated single-center cohort with pathologic T2–T4 gastric cancer, CEUS-GF showed good agreement with pathology (exact accuracy, 84.4%; κ, 0.781) and high discrimination for pathologic T4 disease (AUC, 0.966). CEUS-GF may be a useful adjunct for assessing suspected serosal or extragastric extension before surgery; prospective multicenter paired-comparison studies are needed to define its comparative and incremental value relative to CT and EUS.
Supplementary
The article’s supplementary files as
Acknowledgments
None.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The institutional ethics committee of Shengli Oilfield Central Hospital approved the study protocol (No. YXLL202513202) and waived the requirement for written informed consent because the analysis used de-identified retrospective data and did not alter patient management.
Footnotes
Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0800/rc
Funding: None.
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0800/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0800/dss
References
- 1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63.. 10.3322/caac.21834 [DOI] [PubMed] [Google Scholar]
- 2. Patel AK, Sethi NS, Park H. Gastric Cancer: A Review. JAMA 2026;335:439-50.. 10.1001/jama.2025.20034 [DOI] [PubMed] [Google Scholar]
- 3. Shitara K, Fleitas T, Kawakami H, et al. Pan-Asian adapted ESMO Clinical Practice Guidelines for the diagnosis, treatment and follow-up of patients with gastric cancer. ESMO Open 2024;9:102226.. 10.1016/j.esmoop.2023.102226 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Ungureanu BS, Sacerdotianu VM, Turcu-Stiolica A, et al. Endoscopic Ultrasound vs.Computed Tomography for Gastric Cancer Staging: A Network Meta-Analysis. Diagnostics (Basel) 2021;11:134. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. de Nucci G, Gabbani T, Impellizzeri G, et al. Linear EUS Accuracy in Preoperative Staging of Gastric Cancer: A Retrospective Multicenter Study. Diagnostics (Basel) 2023;13:1842.. 10.3390/diagnostics13111842 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Aslanian HR, Muniraj T, Nagar A, et al. Endoscopic Ultrasound in Cancer Staging. Gastrointest Endosc Clin N Am 2024;34:37-49.. 10.1016/j.giec.2023.09.009 [DOI] [PubMed] [Google Scholar]
- 7. Zhang X, Yao J, Zhang Y, et al. Updated Evaluation of the Diagnostic Performance of Double Contrast-Enhanced Ultrasonography in the Preoperative T Staging of Gastric Cancer: A Meta-Analysis and Systematic Review. Front Oncol 2022;12:844390.. 10.3389/fonc.2022.844390 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Wu LL, Xin JY, Wang JJ, et al. Prospective Comparison of Oral Contrast-Enhanced Transabdominal Ultrasound Imaging With Contrast-Enhanced Computed Tomography in Pre-operative Tumor Staging of Gastric Cancer. Ultrasound Med Biol 2023;49:569-77.. 10.1016/j.ultrasmedbio.2022.10.010 [DOI] [PubMed] [Google Scholar]
- 9. Liang Y, Jing WY, Song J, et al. Clinical application of oral contrast-enhanced ultrasound in evaluating the preoperative T staging of gastric cancer. World J Gastroenterol 2024;30:4439-48.. 10.3748/wjg.v30.i41.4439 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Lv M, Hui X, Yang X, et al. Comparison of the diagnostic accuracy of enhanced-CT and double contrast-enhanced ultrasound for preoperative T-staging of gastric cancer: a meta-analysis. Cancer Imaging 2025;25:48.. 10.1186/s40644-025-00861-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.AJCC. AJCC Cancer Staging Manual. 8th ed. New York: Springer; 2017. [Google Scholar]
- 12. Cohen J. A Coefficient of Agreement for Nominal Scales. Educational and Psychological Measurement 1960;20:37-46. [Google Scholar]
- 13. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159-74. [PubMed] [Google Scholar]
- 14. Demidenko E. Laplace's law of succession estimator and M-statistics. Am Stat 2025;79:311-9.. 10.1080/00031305.2024.2448430 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Ajani JA, D'Amico TA, Bentrem DJ, et al. Gastric Cancer, Version 2.2025, NCCN Clinical Practice Guidelines In Oncology. J Natl Compr Canc Netw 2025;23:169-91.. 10.6004/jnccn.2025.0022 [DOI] [PubMed] [Google Scholar]
- 16. Zhong Y, Xiao YY, Ye JY, et al. Diagnostic efficacy of contrast-enhanced gastric ultrasonography in staging gastric cancer: a meta-analysis. BMC Cancer 2024;24:422.. 10.1186/s12885-024-12210-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Wani AH, Parry AH, Feroz I, et al. Preoperative Staging of Gastric Cancer Using Computed Tomography and Its Correlation with Histopathology with Emphasis on Multi-planar Reformations and Virtual Gastroscopy. J Gastrointest Cancer 2021;52:606-15.. 10.1007/s12029-020-00436-6 [DOI] [PubMed] [Google Scholar]
- 18. Schuster IP, DiMaio CJ. Endoscopic Ultrasound in Gastric Cancer. Surg Clin North Am 2025;105:65-73.. 10.1016/j.suc.2024.07.003 [DOI] [PubMed] [Google Scholar]
- 19. Zhong MY, Jian GL, Ye JY, et al. Clinical characteristics and transabdominal ultrasound features that associated with T3-T4 staging in gastric cancer: A retrospective study. World J Gastrointest Surg 2025;17:105073.. 10.4240/wjgs.v17.i6.105073 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Hirata Y, Agnes A, Prakash LR, et al. Borrmann Type Predicts Response to Preoperative Therapy in Advanced Gastric Cancer. J Gastrointest Cancer 2023;54:882-9.. 10.1007/s12029-022-00880-6 [DOI] [PubMed] [Google Scholar]