This systematic review and meta-analysis is reported in accordance with the Preferred Reporting Items for Systematic Evaluation and Meta-Analysis (PRISMA) guidelines. This systematic evaluation program was registered with the Prospective International Register of Systematic Evaluations (PROSPERO) online database (PROSPERO identifier: CRD42023452140).

We conducted a literature search for English-language articles on clinical studies published from January 1, 2018, to October 1, 2023, using Web of science, EMBASE, PubMed, and the Cochrane Library databases. Non-English articles, unpublished studies, case reports, abstracts, conference abstracts and animal studies were not included. Search terms included “early gastric cancers”, “early-stage gastric cancer”, “Early Detection of gastric cancer”, “Early Detection of gastric cancer”, “gastric carcinoma stomach cancer”, “Intestinal metaplasia”, “IEE”, “linked color imaging “, “linked color imaging endoscopy”, “Image enhanced”, “Image enhanced endoscopy”, “IEE”, “LCI”, “White light”, “White light endoscope”, “white light endomicroscopy”. A complete strategy for each database can be found in the Supplementary Materials ( Supplementary Document S1 ).

We implemented predefined broad eligibility criteria and included prospective and retrospective studies to validate the diagnostic performance of WLI and LCI for EGC and gastrointestinal metaplasia when compared with gold standard biopsy histopathology.

The included criteria are as follows: (1) all patients underwent LCI and WLI; (2) true positive (TP), false positive (FP), true negative (TN), and false negative (FN) values, or values that could be calculated based on sensitivity (Se) and specificity (Sp), were reported. (3) Detection rates were reported or could be calculated from relevant data (4) The literature stated that there was a “gold standard” (pathologic examination) for confirmation of the diagnosis. Exclusion Criteria: (1) Studies published in the form of reviews, case studies, and conference papers were excluded. (2) Literature that could not specify the number of cases and lesions. (3) Articles using the same data set were considered duplicates. (4) Literature that did not specify a “gold standard”.

All acquired studies were transferred to EndNote X9 where duplicates were identified and eventually eliminated. To ensure that relevant studies were not overlooked, the reference lists of all eligible articles were manually reviewed.

Two reviewers (HuiDan, XingxuZhou) screened the titles and abstracts of all articles identified in the initial literature search and then extracted data from selected articles. Disagreements were resolved by consensus. For the included articles, the full text and its associated references were reviewed in their entirety.

We extracted data from each eligible study, including first author, year of publication, country, included studies, number of lesions, sensitivity, specificity, PPV, NPV, and detection rate. This was done by two independent reviewers, with third-party input if required. All data were summarized and compiled into a table accessible to all authors.

The methodological quality of the selected diagnostic accuracy studies was assessed using the Quality Assessment Tool for Diagnostic Accuracy Studies (QUADAS-2) (Whiting et al., 2011), which was completed independently by two reviewers, with group discussion as necessary. The QUADAS-2 Quality Assessment Scale consists of four sections: (1) case selection, (2) tests to be evaluated, (3) gold standard, (4) case flow and progress. The QUADAS-2 quality assessment scale consists of: (1) selection of cases, (2) trials to be evaluated, (3) gold standard, (4) case flow and progression, and these four sections contain 10 items. The first 3 sections are evaluated in terms of clinical applicability, and the 10 items are evaluated in terms of “yes”, “no”, “no”, “no”, “no” and “no”. All 10 items were judged according to the criteria of “yes”, “no”, and “unclear”, and the corresponding risk of bias was “low”, “no”, “unclear”, and “low”. The corresponding risk of bias was “low”, “high”, “no”, and “unclear”, “high” and “unclear”, respectively. Among them, “unclear” was mainly used when the included literature did not provide detailed content and complete data, and it was difficult for the researchers to make a judgment. The risk of bias was considered low only if all items in a section were “yes”. If one item was “no”, the section might be at risk of bias; the quality evaluation chart, risk of bias and applicability evaluation summary chart were drawn by Revman 5.4 software.

The extracted data were analyzed using Review Manager 5.4 software and Stata 15 software.

For the detection rate, data were analyzed using Review Manager 5.4 software. Count data were expressed as 95% confidence interval (95% CI), and Odds risk (OR) was used to test the count data, comparing the detection rate of EGC and the detection rate of gastrointestinal metaplasia between LCI and WLI. Heterogeneity between included studies was analyzed by χ² test and described by I² value. Low heterogeneity between studies was suggested when I² >25%, moderate heterogeneity when 50%<I²<75%, and high heterogeneity when I² >75%, and substantial heterogeneity was generally considered to exist for I² >50%. A random-effects model was used when I² >50% and a fixed-effects model when I² <50% (21). P < 0.05 was considered statistically significant. The symmetry of the funnel plot was used to assess publication bias, and a forest plot was used to present the results of data analysis for each group. Sensitivity analysis was then performed using Stata15 software to determine the effect of individual studies on the overall statistical difference, while the merging of I² values of the remaining literature was performed after excluding the included literature one by one. If the total I² value of the remaining studies was significantly lower after the exclusion of a single study compared with that before the exclusion, it means that this literature may be the source of heterogeneity of the articles, and the reasons for the heterogeneity need to be further discussed.

For specificity and sensitivity of LCI and WLI, datas were analyzed using Stata 15 software. The Spearman correlation coefficient was used to assess the threshold effect (a strong positive correlation of the Spearman coefficient suggests the possibility of a threshold effect) (22). The degree of inter-study heterogeneity was examined using the Higgins Inconsistency Index (I²), with a value of the I² of less than 25% being regarded as low, a value of the I² of 25% to 50% being regarded as moderate, and a value of the I² of 50% to 75% being regarded as high. Heterogeneity between studies was considered high when the I² value was between 50% and 75%, and very high when the I² value was above 75%²⁵. A binary generalized mixed bivariate model was used to estimate the magnitude of the combined effects, including: Summarized Sensitivity (SSEN), Specificity (SSPE), and Specificity (SSPE). Specificity (SSPE), Summarized Positive Likelihood Ratio (SPLR), Summarized Negative Likelihood Ratio (SNLR), and Diagnostic Odd Ratio (SNLR) and Diagnostic Odd Ratio (DOR). Mori diagrams and Summary Receiver Operating Characteristics (SROC) curves were plotted and Area Under Summary Receiver Operating Characteristics (AUSROC) was calculated. Publication bias was explored by Deeks’ funnel plot, and we plotted Deek’s funnel plot asymmetry test for LCI and WLI respectively. If the funnel plot showed a result of P≥0.05, it indicated that there was no publication bias; if the funnel plot showed a result of P<0.05, it indicated that there was publication bias. Fagan plots were then plotted to reflect the diagnostic value of the diagnostic methods, and the posterior probability was calculated by setting the a priori probability.

Figure 1 illustrates the PRISMA flowchart, which demonstrates the process of study selection. Our exhaustive database search initially yielded 1010 articles. Of these, 328 were eliminated through EndNote and 682 were screened for titles and abstracts. Following title and abstract screening, 660 publications were excluded, leaving only 22 full-text screens. The full text of these 22 records was retrieved and reviewed for eligibility. Eleven articles were excluded for various reasons summarized in Figure 1 , resulting in a total of 11 documents included in the analysis.

Tables 1 , 2 show the extracted datasets of the included articles. Briefly, they were published between 2019 and 2023, all in Asian countries, with the majority from Japan (n=7), China (n=3), and Singapore (n=1).11 studies (23–33), with 3 articles retrospectively included Patients were patients with pre-existing gastrointestinal tumors and were evaluated by a specialists to evaluate past samples/images (Toshihisa 2019; Masayuki 2023; Tsevelnorov 2022);. The remaining 8 articles were prospective studies (Ken Haruma 2022; MinMin 2022; Minoru Yamaoka 2019; Shoko Ono 2021; Honglei Cheng 2019; Jiang Zhang Xiu 2021; Shoko Ono 2018; Clement 20) All articles Confirmation of the diagnosis was based on pathological findings.11 articles focused on comparing the difference between LCI and WLI in terms of detection rate, specificity, and sensitivity for early-stage cancer and gastrointestinal chemosis.

Seven articles evaluated the detection rates of Linked Color Imaging (LCI) and White Light Imaging (WLI) in early-stage cancer, while four articles assessed the detection rates of LCI and WLI in gastrointestinal metaplasia. Additionally, four articles examined the specificity and sensitivity of LCI in detecting early cancers, while another four articles assessed the specificity and sensitivity of both LCI and WLI in detecting gastrointestinal metaplasia. Ultimately, seven articles (n=7) were included in the analysis of detection rates for early cancers, and four articles (n=4) were included in the analysis of specificity and sensitivity. Furthermore, gastrointestinal metaplasia was analyzed in four articles pertaining to detection rates, specificity, and sensitivity.

The assessment indicates that most of the included studies did not pre-specify a threshold, leading to inconclusive answers regarding this aspect. Additionally, while the risk of bias and clinical applicability were evaluated, the overall assessment did not suggest a high-risk evaluation ( Figure 2 ).

White Light Imaging (WLI) technology has been instrumental in detecting gastric cancer and its precancerous lesions, significantly contributing to the diagnosis and treatment of gastric diseases since its inception. However, due to the subtle morphological manifestations and color similarities between early-stage gastric cancer (EGC) and its precancerous lesions, there is a risk of missed diagnosis, even for endoscopists who are well-versed in endoscopic practices. This potential oversight can lead to delayed diagnosis, emphasizing the need to enhance detection rates.

Historically, pigment endoscopy was employed to increase detection rates, offering superior contrast and sensitivity compared to white light imaging. However, this method presented challenges such as complex stain preparation, mucosal irritation, allergic reactions in some patients, time consumption, uneven staining, and difficulty in determining tumor depth, rendering it unsuitable for routine screening (34).

The advent of Linked Color Imaging (LCI) endoscopy has spurred continued exploration by endoscopists worldwide. Studies, including a multicenter large-sample study by Minmin et al. (24) have indicated that LCI can effectively identify EGCs during routine endoscopy by enhancing color differentiation of gastric mucosa, thereby improving detection rates. Additionally, research by Jiangzhang Xiu et al (33). suggests that LCI demonstrates higher diagnostic accuracy for intestinal metaplasia compared to WLI.

To further elucidate the potential benefits of LCI in enhancing detection rates of both EGC and GIM, we conducted a systematic meta-analysis encompassing 11 high-quality randomized controlled trials (RCTs). The analysis included a total of 6,492 patients in the EGC group and 1,344 cases in the GIM group, providing a comprehensive evaluation of whether LCI can improve detection rates for both conditions.

To the best of our knowledge, this meta-analysis represents the first comprehensive assessment of the detection rate of Linked Color Imaging (LCI) for Early Gastric Cancer (EGC) and gastrointestinal metaplasia (GIM). In this systematic review and meta-analysis, we scrutinized the accuracy of LCI and White Light Imaging (WLI) in diagnosing EGC and GIM, revealing that LCI outperformed WLI in both conditions.

In the diagnosis of EGC, LCI exhibited a detection rate of 85.75%, whereas WLI showed a detection rate of 56.70%. Similarly, in diagnosing gastrointestinal metaplasia, LCI demonstrated a detection rate of 88.89%, whereas WLI yielded a detection rate of 40.12%. Moreover, LCI displayed higher specificity and sensitivity in diagnosing both EGC and gastrointestinal metaplasia. The diagnostic efficacy of LCI for EGC was characterized by a sensitivity (SEN) of 0.80 (95% CI 0.71-0.86) and specificity (SPE) of 0.82 (95% CI 0.63-0.92), with a Positive Likelihood Ratio (PLR) of 4.4 (95% CI 2.1-4.9) and a Negative Likelihood Ratio (NLR) of 0.25 (95% CI 0.18-0.33). Similarly, for the diagnosis of gastrointestinal metaplasia, LCI demonstrated a sensitivity (SEN) of 0.87 (95% CI 0.81-0.92) and a specificity (SPE) of 0.85 (95% CI 0.77-0.91), with a PLR of 6.0 (95% CI 3.7-9.6) and an NLR of 0.15 (95% CI 0.10-0.23).In contrast, the sensitivity and specificity of WLI for diagnosing gastrointestinal metaplasia were notably lower, with a SEN of 0.39 (95% CI 0.22-0.60) and a SPE of 0.89 (95% CI 0.56-0.98), alongside a PLR of 3.7 (95% CI 0.8-17.0) and an NLR of 0.68 (95% CI 0.51-0.90).

These findings underscore the superior sensitivity and specificity of Linked Color Imaging (LCI) in diagnosing both Early Gastric Cancer (EGC) and gastrointestinal metaplasia (GIM), highlighting its enhanced diagnostic value compared to White Light Imaging (WLI). Notably, the results of this study are consistent with numerous previously published clinical studies. For instance, two large-scale, multicenter, randomized controlled trials conducted in Japan demonstrated the utility of LCI in accurately identifying gastric tumor lesions (23, 24). Additionally, studies by Haruma K (23) and Khurelbaatar T et al (27). revealed that both ultra-thin and standard LCI exhibited superior diagnostic efficacy over WLI for diagnosing EGC. Further corroborating these findings, Higashino M et al (28). illustrated that LCI surpassed WLI in sensitivity for EGC screening, particularly among novice and trainee endoscopists. Moreover, Ono S et al (26). found LCI to outperform WLI in identifying early gastric cancers in patients with atrophic gastritis. Importantly, LCI also demonstrated commendable efficacy compared to WLI in detecting GIM across several studies. Collectively, these findings suggest that LCI holds a significant advantage over WLI in the detection of both EGC and GIM, highlighting its potential as a superior diagnostic tool in clinical practice.

LCI, developed by Fuji in Japan in 2012, is a blue laser endoscope known as LASEREO. It emits four types of light in the pre-processing module: amber, green, white (450nm), and violet (410nm). The violet light, with its narrow wavelength, can penetrate only the surface of the mucosa. When absorbed by the surface hemoglobin, it is not reflected, but in deeper blood vessels, it is reflected. This unique property enables the differentiation of various lesions based on differences in the depth of blood vessels. By magnifying the color difference, LCI enhances the visual contrast, making whites appear whiter and reds appear redder (35). Liu Yan et al. established a color-microstructure-vessel (CMV) model based on the microscopic features of color, microstructure, and vascularity of the gastrointestinal (GI) mucosa. This model delineates three color categories: typical red (type 1), purplish-red (type 2), and purplish-ringed-yellow-red (type 3). In LCI, intestinal metaplasia appears purple, inflammation appears red, and normal mucosa or tumors appear yellow. Chronic non-atrophic gastritis, characterized by non-specific inflammation of the distal stomach, typically appears red on LCI. Gastric cancer, with its complex pathological changes, exhibits a varied color spectrum, typically showing purple ringed yellow-red hues (36).

LCI has been demonstrated to be a superior Image-Enhanced Endoscopy (IEE) technique compared to WLI because it offers better visualization of red and purple changes. This enhanced visibility makes the identification and detection of EGC or GIM easier and more accurate (30, 37–39).

The robustness of our findings is supported by several factors. Firstly, the substantial number of cases included in this study, totaling 7,836, ensures the reliability and representativeness of the results. Secondly, the study populations were exclusively composed of Asian individuals, minimizing potential biases related to racial classification. Additionally, all studies utilized pathological findings as the definitive criterion for judgment, ensuring consistency and accuracy in the evaluation process. Furthermore, eight of the included studies were prospective, enhancing the standardization of Randomized Controlled Trial (RCT) protocols and providing a more realistic reflection of clinical gastroenteroscopy practices. This ensures that our findings are applicable not only at a theoretical level but also in real-world clinical settings.

Several limitations should be acknowledged in our study. Firstly, publication bias may have influenced the results, as studies reporting positive findings regarding the efficacy of LCI in detecting EGC and precancerous lesions might be more likely to be published. Conversely, studies with negative results may be underreported, potentially biasing the overall findings. Secondly, inherent heterogeneity among the included studies could have impacted the outcomes. Factors such as variations in bowel preparation, timing of gastroscopy, differences in endoscopists’ experience levels, and the inclusion criteria for patients (including both general patients and those at high risk of gastric cancer) may have introduced inconsistencies and affected the accuracy of our meta-analysis. Moreover, due to data limitations, we were unable to conduct subgroup analyses to assess diagnostic tests specifically for EGC and gastrointestinal metaplasia separately. Early gastric cancer may depend on the use of magnifying endoscopy and thin endoscopy, but due to the limited data in this paper we were unable to perform subgroup analyses to clarify its impact. Lastly, our study’s endpoint, the leakage rate, was determined based on the final results of two endoscopies, which may have limitations as it does not guarantee 100% accuracy in detecting lesions without leakage.