Synergistic applications of artificial intelligence and organoid technology in gastric precancerous lesion research: Mechanisms, translation, and challenges

Abstract

Gastric precancerous lesions (GPLs) are critical stages in gastric carcinogenesis, where early and precise identification and intervention are pivotal for improving patient prognosis. Recent advancements in artificial intelligence (AI) and organoid technology have revolutionized GPL research, with their synergistic applications progressively overcoming limitations of traditional methodologies. This review systematically summarizes the latest progress in AI and organoid technology for GPLs, focusing on mechanistic exploration, diagnostic optimization, and therapeutic innovation. We highlight how AI enhances diagnostic accuracy in endoscopy and pathology, while organoids provide unparalleled models for studying disease progression and drug screening. Critically, we discuss the pioneering integration of both technologies, wherein AI analyzes dynamic organoid phenotypes and fuses multimodal data for risk prediction. Current challenges hindering clinical translation, such as data standardization and model interpretability, are critically examined. We conclude that the future of GPL management lies in harnessing this synergy through interdisciplinary collaboration, which is poised to bridge the bench-to-bedside gap and usher in a new paradigm of mechanism-based precision prevention for gastric cancer.

Key Words: Gastric precancerous lesions; Artificial intelligence; Organoids; Integrative research; Precision medicine

Core Tip: We propose a transformative paradigm: Using artificial intelligence to decipher dynamic organoid phenotypes for forecasting gastric cancer (GC) risk. This review details how this synergy, which integrates high-throughput organoid screening with multimodal clinical data, unlocks the “black box” of disease progression. It empowers a shift from reactive diagnosis to proactive, mechanism-based precision prevention, ultimately pioneering a new era where GC is preemptively intercepted rather than treated.

INTRODUCTION

Gastric cancer (GC) is a common malignant tumor worldwide, with its incidence and mortality rates ranking among the highest, posing a serious threat to human health[1-5]. Its pathogenesis follows a multistep process, often adhering to the “Correa cascade” model, which progresses from chronic gastritis, atrophy, and intestinal metaplasia (IM) to dysplasia and ultimately to carcinoma[6]. Therefore, early identification, accurate diagnosis, and effective intervention of gastric precancerous lesions (GPLs) [e.g., chronic atrophic gastritis (CAG), IM] are critical strategies for reducing the incidence and mortality of GC[7,8]. The prognosis of GC is closely tied to the timing of diagnosis. If early GC (EGC) is diagnosed and treated promptly, the patient’s 5-year overall survival rate can be significantly improved[3,9]. However, due to atypical or absent early symptoms, most patients are diagnosed at advanced stages, resulting in suboptimal treatment outcomes[10]. Consequently, advancing mechanistic research to enhance the detection rate of precancerous lesions and improve the efficacy of treatments is crucial for improving patient prognosis[5,8,9].

Traditional diagnosis and therapeutic decision-making for GC primarily rely on endoscopic examination, imaging, and pathological evaluation[8]. However, these methods are constrained by physician experience, interobserver variability, and the complexity of lesion morphology, which can lead to insufficient diagnostic accuracy[1,5,11]. Upper gastrointestinal endoscopy combined with biopsy is the current gold standard for diagnosing GPLs[8]. Nonetheless, conventional biopsy has limitations such as sampling error and interobserver variability[7]. Particularly in the early stages, lesions can be highly occult and difficult to detect with the naked eye[11-13].

In recent years, two cutting-edge technologies-artificial intelligence (AI) and organoid models-have brought revolutionary changes to GC research and clinical management. AI, particularly machine learning and deep learning algorithms, can assist clinical decision-making by analyzing vast, multidimensional medical data (e.g., endoscopic images, pathological slides, imaging data, and genomic information)[1,14]. Numerous studies have confirmed AI’s significant potential in improving the early detection rate, diagnostic accuracy and efficiency, TNM staging, treatment response prediction, and prognosis assessment of GC[1,2,4,9,14].

Simultaneously, patient-derived cancer organoids, as a novel three-dimensional (3D) cell culture technology, can highly simulate the biological characteristics of the original tumor, including genetic, phenotypic, and behavioral features[15]. This makes organoids an ideal model for studying tumorigenesis mechanisms, drug screening, and predicting individualized treatment responses[16-18].

Although both technologies have made significant progress independently in GC research, most studies have primarily focused on two aspects: First, the application of AI in the imaging and pathological diagnosis of GC and its precancerous lesions; and second, the role of organoids in simulating GC and its microenvironment, elucidating the mechanisms of immune escape in GC, and guiding the treatment of advanced GC. Research combining both technologies for application in GPLs is still in its exploratory stages. This constitutes the primary significance of this review: To venture into this uncharted territory of interdisciplinary integration by synthesizing existing evidence, analyzing the current applications of AI and organoids in exploring the mechanisms and clinical management of GC and its precancerous lesions, and on this basis, discussing potential innovative directions for their combination. This aims to propel the innovation of strategies for early screening, risk stratification, and intervention. This integrative perspective can provide a novel paradigm for the precise prevention and control of GPLs and represents an important, highly regarded topic in the field of precision oncology.

APPLICATION AND VALUE OF AI IN THE DIAGNOSIS OF GC AND PRECANCEROUS LESIONS

In recent years, AI, particularly computer-aided diagnosis systems based on deep learning and convolutional neural networks (CNN), has presented new opportunities for improving the accuracy and consistency of digestive disease diagnosis[9,19].

AI demonstrates high accuracy in diagnosing GC and precancerous lesions

Numerous studies have confirmed that AI systems can assist endoscopists in identifying EGC and its precancerous lesions. A 2022 systematic review, which included 12 retrospective case-control studies (totaling 11685 cases), reported that AI achieved a pooled sensitivity of 0.86 [95% confidence interval (CI): 0.75-0.92], a specificity of 0.90 (95%CI: 0.84-0.93), and an area under the curve (AUC) of 0.94 for EGC diagnosis, indicating high diagnostic performance[20]. A 2020 meta-analysis of 19 studies evaluated AI's performance in detecting upper gastrointestinal tumors (including gastric adenocarcinoma) and found an overall sensitivity of 90%, specificity of 89%, and an AUC of 0.95[21]. Furthermore, another 2020 systematic review and meta-analysis (incorporating 23 studies and 969318 images) discovered that AI achieved an AUC of 0.96 for detecting gastric neoplastic lesions and outperformed endoscopists (AUC 0.98 vs 0.87, P < 0.001)[22]. These studies further substantiate the supportive role of AI in the diagnosis of EGC and demonstrate its effectiveness in assisting physicians, thereby reducing missed diagnoses due to inexperience or fatigue[5].

AI also excels in diagnosing precancerous lesions. Substantial research confirms that CNN-based AI systems exhibit high accuracy in the endoscopic diagnosis of CAG and IM, with performance often matching or surpassing that of endoscopists, particularly providing significant aid to less-experienced physicians. A 2022 systematic review and meta-analysis, which included 4 studies on GPLs, reported an AI diagnostic accuracy of 90.3%[23]. A 2023 systematic review and meta-analysis of 8 studies, involving 25216 patients and over 90000 images, showed that AI achieved a pooled sensitivity of 94%, a specificity of 96%, and a summary AUC of 0.98 for identifying CAG, with diagnostic accuracy significantly higher than that of endoscopists[24]. Multiple independent studies have reached similar conclusions. A 2020 study utilizing 5470 gastric antrum images to train a CNN model reported accuracies, sensitivities, and specificities of 0.942, 0.945, and 0.940, respectively, for diagnosing CAG, all exceeding the results from three expert endoscopists[25]. Another 2023 study developed a GAM-Efficient Net model that achieved accuracies of 93.5% and 92.37% on external image and video test sets, respectively, also outperforming endoscopists[26]. A 2020 study developed an intelligent diagnostic system for gastric IM that achieved an AUC of 0.928, with a sensitivity, specificity, and accuracy of 91.9%, 86.0%, and 88.8%, respectively[27]. A 2021 study developed an AI-DDx model for the differential diagnosis of gastric mucosal lesions; its performance was comparable to experts [area under the receiver operating characteristic curve (AUROC) 0.86 vs 0.89, P = 0.12] but superior to novices (AUROC = 0.82, P = 0.01) and intermediate endoscopists (AUROC = 0.84, P = 0.02)[28].

AI also demonstrates excellent performance in simultaneously identifying CAG and IM. A 2021 multicenter study, using 7037 endoscopic images from 14 hospitals in China, showed that a CNN model achieved an AUC of 0.98 and an accuracy of 96.4% for identifying CAG, and an AUC of 0.99 with an accuracy of 97.6% for identifying IM[29]. These findings were cited and endorsed by the “Chinese Guidelines for the Diagnosis and Treatment of Chronic Gastritis (2022, Shanghai)”, which states that CNN models based on white-light endoscopic images possess high diagnostic accuracy for atrophic gastritis and IM[30]. Additionally, a novel deep learning method proposed in a 2023 study was tested on a dataset of 21420 images, achieving accuracies of 97.12% for detecting atrophic gastritis and 99.18% for detecting IM[31]. A 2021 multicenter diagnostic study developed a deep CNN system named ENDOANGEL to evaluate AI's capability in diagnosing gastric atrophy (GA) and IM using image-enhanced endoscopy. The study incorporated 6250 endoscopic images and 98 videos from five hospitals. Results indicated that in multicenter external test sets, the diagnostic accuracies for GA and IM were 86.4% and 85.9%, respectively, comparable to expert performance and superior to non-experts[32]. A 2025 prospective single-center clinical trial assessed the impact of an AI system on the detection of precancerous lesions, revealing that AI assistance significantly increased the detection rates of IM (14.23% vs 9.15%) and atrophy (22.76% vs 17.28%), with the effect being more pronounced among junior physicians[33].

Collectively, these studies indicate that AI systems can capture subtle image features imperceptible to the human eye through deep learning, enabling performance in diagnosing CAG and IM that meets or exceeds that of human experts. This suggests AI has the potential to be a powerful tool for screening high-risk populations, providing a reliable auxiliary diagnostic means for clinical practice[30], and effectively assisting endoscopists in identifying GPLs and EGC, reducing missed diagnoses, and improving diagnostic quality[5,34,35].

Furthermore, AI is capable not only of detecting lesions but also of analyzing their characteristics. Research shows that AI systems can be used to predict the depth of GC invasion, determine differentiation type, identify lesion boundaries, differentiate between benign and malignant lesions, and distinguish confusing neoplastic lesions. This capability for detailed diagnosis and prediction directly influences treatment decisions, providing predictive evidence to support the choice of treatment strategies, such as endoscopic therapy or surgery[9]. A 2021 study developed an AI-ID model for assessing the invasion depth of GC; its performance was significantly superior to expert-conducted endoscopic ultrasound (EUS) (AUROC 0.73 vs 0.56, P < 0.001)[28]. A 2020 study developed a novel AI diagnostic system (A-CNN) that excelled in distinguishing GC from gastric ulcers, demonstrating a sensitivity of 99.0%, a specificity of 93.3%, and an overall classification accuracy of 95.9% for GC[36]. A 2024 multicenter diagnostic study showed that a real-time AI-assisted EUS diagnosis system achieved an AUC of 0.948 and an accuracy of 91.7% in differentiating gastrointestinal stromal tumors from leiomyomas, significantly outperforming endoscopists[37].

Beyond endoscopic images, AI can also be applied to pathological diagnosis. In the field of digital pathology, by analyzing digital whole-slide images, AI can assist in detecting micrometastases, quantitatively scoring immunohistochemical markers (e.g., HER2, PD-L1), and analyzing the tumor microenvironment (TME), thereby enhancing diagnostic precision and reproducibility[38,39]. A 2021 study employed an AI algorithm to analyze macroscopic images of GC surgical specimens, localizing cancerous foci with a mean average precision of 95.90% in the test set and predicting lymph node metastasis with an accuracy of 75.00%, demonstrating its potential for rapid lesion localization and improving pathological processing efficiency[40]. Another 2023 study showed that an AI system had only a 7.6% discrepancy with pathologists in classifying IM scores and could identify minute lesions missed by pathologists, highlighting its potential for achieving global standardized assessment[41]. Table 1 summarizes the applications of AI in diagnosing GC and precancerous lesions.

Table 1 Diagnostic performance of artificial intelligence systems for gastric cancer and precancerous lesions.

No.	Research focus	Ref.	Modality	Study design	Sample size/dataset	AI model/system	Gold standard	Key performance metrics	Results	Comparison with endoscopists
1	EGC diagnosis	Chen et al[20]	WLE, NBI	Systematic review (12 studies)	11685 cases	Various	Pathology	Pooled sensitivity, specificity, AUC	Sensitivity: 0.86 (95%CI: 0.75-0.92); specificity: 0.90 (95%CI: 0.84-0.93); AUC: 0.94	Not compared
2	Upper GI tumors	Arribas et al[21]	NBI	Meta-analysis (19 studies)	Not specified	Various	Pathology	Overall sensitivity, specificity, AUC	Sensitivity: 90%; specificity: 89%; AUC: 0.95	Not compared
3	Gastric neoplasia	Lui et al[22]	WLE, NBI	Systematic review and meta-analysis (23 studies)	969318 images	Various	Pathology	AUC	AUC: 0.96	Superior (AUC 0.98 vs 0.87, P < 0.001)
4	GPLs diagnosis	Dilaghi et al[23]	Not Specified	Systematic review and meta-analysis (4 studies)	Not specified	Various	Pathology	Accuracy	Accuracy: 90.3%	Not compared
5	CAG diagnosis	Shi et al[24]	Not specified	Systematic review and meta-analysis (8 studies)	25216 patients, > 90000 images	Various	Pathology	Pooled sensitivity, specificity, AUC	Sensitivity: 94%; specificity: 96%; AUC: 0.98	Significantly higher accuracy
6	CAG diagnosis	Zhang et al[25]	Not specified	Diagnostic study	5470 antral images	CNN	Pathology	Accuracy, sensitivity, specificity	Accuracy: 0.942; sensitivity: 0.945; specificity: 0.940	Exceeded three experts
7	CAG diagnosis	Shi et al[26]	Not specified	Diagnostic study	Not specified	GAM-efficient net	Pathology	Accuracy	External image: 93.5%; video: 92.37%	Outperformed endoscopists
8	GIM diagnosis	Yan et al[27]	Not specified	Diagnostic study	Not specified	Intelligent diagnostic system	Pathology	AUC, sensitivity, specificity, accuracy	AUC: 0.928; sensitivity: 91.9%; specificity: 86.0%; Accuracy: 88.8%	Not compared
9	Mucosal lesion DDx	Nam et al[28]	Not specified	Diagnostic study	Not specified	AI-DDx	Pathology	AUROC	AUROC: 0.86	Comparable to experts (0.89, P = 0.12); Superior to novices and intermediates
10	CAG and IM diagnosis	Lin et al[29]	WLE	Multicenter diagnostic study	7037 images (14 hospitals)	CNN	Pathology	AUC, Accuracy	CAG: AUC 0.98, Acc 96.4%; IM: AUC 0.99, Acc 97.6%	Not compared
11	Atrophy and IM detection	Yang et al[31]	WLE, LCI	Diagnostic study	21420 images	Novel DL method	Pathology	Accuracy	Atrophy: 97.12%; IM: 99.18%	Not compared
12	GA and IM diagnosis	Xu et al[32]	Image-enhanced endoscopy	Multicenter diagnostic study	6250 images, 98 videos (5 hospitals)	ENDOANGEL (DCNN)	Pathology	Accuracy	GA: 86.4%; IM: 85.9%	Comparable to experts; Superior to non-experts
13	Precursor detection	Xu et al[33]	Not specified	Prospective single-center clinical trial	Not specified	Not specified	Pathology	Detection Rate	IM: 14.23% vs 9.15%; atrophy: 22.76% vs 17.28%	Effect more pronounced in junior physicians
14	Invasion depth	Nam et al[28]	EUS	Diagnostic study	Not specified	AI-ID	Post-operative histology	AUROC	AUROC: 0.73	Superior to EUS experts (0.56, P < 0.001)
15	Cancer vs ulcer	Namikawa et al[36]	Not specified	Diagnostic study	Not specified	A-CNN	Pathology	Sensitivity, specificity, accuracy	Sensitivity: 99.0%; specificity: 93.3%; accuracy: 95.9%	Not compared
16	GISTs vs leiomyomas	Dong et al[37]	EUS	Multicenter diagnostic study	Not specified	Real-time AI-assisted EUS system	Pathology/histology	AUC, accuracy	AUC: 0.948; accuracy: 91.7%	Significantly outperformed
17	Pathology analysis	Yang et al[40]	Macroscopic specimen	Diagnostic study	Gastric cancer surgery specimens	AI algorithm	Histology	mAP, Accuracy	Lesion localization mAP: 95.90%; LN metastasis prediction Acc: 75.00%	Not compared
18	IM scoring	Iwaya et al[41]	Pathology slides	Diagnostic study	Not specified	AI system	Expert pathologist	Scoring difference	Difference with pathologists: 7.6%	Identified missed foci by pathologists

AI: Artificial intelligence; AUC: Area under the curve; CAG: Chronic atrophic gastritis; CNN: Convolutional neural network; DCNN: Deep CNN; DDx: Differential diagnosis; DL: Deep learning; EGC: Early gastric cancer; EUS: Endoscopic ultrasound; GA: Gastric atrophy; GIM: Gastric intestinal metaplasia; GISTs: Gastrointestinal stromal tumors; GPLs: Gastric precancerous lesions; IM: Intestinal metaplasia; LN: Lymph node; mAP: Mean average precision; WLE: White light endoscopy; NBI: Narrow-band imaging; AUROC: Area under the receiver operating characteristic curve; LCI: Linked color imaging endoscopy; GI: Gastrointestinal; CI: Confidence interval; Acc: Accuracy.

AI applications effectively reduce inter-observer variability and improve diagnostic consistency

AI not only exhibits high diagnostic performance but also significantly serves as an auxiliary tool to enhance the capabilities of endoscopists at all levels. By delivering objective and reproducible analytical results, AI-assisted diagnostic systems substantially reduce diagnostic discrepancies arising from subjective factors like physician experience and fatigue, thereby improving diagnostic consistency.

A 2022 study developed a system (ENDOANGEL-LA) for diagnosing EGC using magnifying image-enhanced endoscopy. Its accuracy (88.76%) matched that of experts but was significantly higher than novices (71.63%). When assisted by this system, novice accuracy markedly improved to 87.45%[42]. Similarly, a 2023 study on a real-time interpretable AI system (ENDOANGEL-ED) demonstrated that AI assistance improved endoscopists' diagnostic accuracy from 70.61% to 79.63% (P < 0.001)[43]. This evidence indicates that AI systems can compensate for the shortcomings of less-experienced physicians and reduce inter-observer diagnostic variability[9]. Another study confirmed that AI assistance significantly improves the diagnostic accuracy of both novice and expert endoscopists for upper gastrointestinal diseases[44]. These findings collectively demonstrate that AI serves as a standardized “second opinion”, effectively bridging the skill gap among physicians of different experience levels and enhancing overall diagnostic consistency.

Some studies have directly compared AI model diagnoses with those of endoscopists. A 2023 prospective nested case-control study directly compared the agreement between an AI model and endoscopists in diagnosing CAG. Involving 1306 patients, the results showed the AI model’s agreement with the gold standard (pathological diagnosis) was substantially higher than that of endoscopists. The Kappa value improved dramatically from 0.291 (fair agreement) for endoscopists to 0.816 (almost perfect agreement) for the AI model. Simultaneously, the AI model’s accuracy (89.89% vs 68.89%), sensitivity (89.31% vs 67.56%), and specificity (90.46% vs 70.23%) comprehensively surpassed those of the endoscopists[45].

In pathological diagnosis, AI similarly reduces subjectivity, improves diagnostic consistency and accuracy, and minimizes interpretation discrepancies among pathologists[38]. A 2023 study developed a semi-supervised deep learning algorithm named GasMIL for diagnosing and grading atrophy and IM in pathological images. In an observer study involving 10 pathologists, GasMIL’s diagnostic performance (AUC 0.953) exceeded that of all pathologists. Furthermore, when assisted by GasMIL, the pathologists showed significant improvements in diagnostic AUC, sensitivity, and weighted kappa values[46].

These studies collectively demonstrate that AI technology, leveraging its powerful image recognition and pattern classification capabilities, provides objective and efficient assistance in diagnosing GPLs, thereby comprehensively enhancing diagnostic quality and efficiency. Table 2 summarizes the role of AI in mitigating inter-observer variability and enhancing diagnostic consistency.

Table 2 Role of artificial intelligence in reducing inter-observer variability and improving diagnostic consistency.

No.	Research focus	Ref.	Modality	Study design	Sample size/dataset	AI model/system	Key performance metrics	Results without AI	Results with AI assistance	Key finding
1	EGC diagnosis with ME	Li et al[42]	Magnifying image-enhanced endoscopy	Diagnostic study	Not specified	ENDOANGEL-LA	Accuracy	Novice: 71.63%	Novice: 87.45%	AI assistance bridged the gap between novices and experts
2	Real-time AI assistance	Dong et al[43]	WLE	Diagnostic study	Not specified	ENDOANGEL-ED	Accuracy	Endoscopists: 70.61%	Endoscopists: 79.63% (P < 0.001)	AI significantly improved endoscopist diagnostic accuracy
3	CAG diagnosis	Zhao et al[45]	Not specified	Prospective nested case-control	1306 patients	Not specified	Kappa, accuracy, sensitivity, specificity	Endoscopists' kappa: 0.291; Acc: 68.89%; Sens: 67.56%; Spec: 70.23%	AI kappa: 0.816; Acc: 89.89%; Sens: 89.31%; Spec: 90.46%	AI agreement with pathology was substantially higher
4	Pathology Dx (atrophy/IM)	Fang et al[46]	Pathology slides	Observer study (10 pathologists)	Not specified	GasMIL (SDL algorithm)	AUC, weighted kappa	Pathologists’ performance (baseline)	Pathologists’ performance significantly improved	AI assistance improved pathologists’ diagnostic metrics

“Not Specified” indicates that the information was not explicitly mentioned in the provided text. AI: Artificial intelligence; CAG: Chronic atrophic gastritis; Dx: Diagnosis; IM: Intestinal metaplasia; ME: Magnifying endoscopy; SDL: Semi-supervised deep learning; Acc: Accuracy; Sens: Sensitivity; Spec: Specificity; AUC: Area under the curves; WLE: White light endoscope.

Limitations and controversies surrounding the integration of AI into the diagnosis of GPLs

Despite its promising prospects, the clinical application of AI faces challenges such as data standardization, algorithm interpretability, multicenter validation, and regulatory approval[1], which also hinder its widespread clinical adoption.

Firstly, model generalizability is a primary concern. Multiple reviews point out that most current studies are single-center or regional, utilizing training datasets typically derived from a single center that are not publicly available. This may lead to degraded performance when models are applied across different medical institutions, endoscopic equipment, or populations[47]. A 2023 external validation study applied an AI model trained on other healthcare systems to a Singaporean population, comparing its diagnostic capabilities against 11 endoscopists (5 experts and 6 non-experts) on 300 endoscopic images. The results showed that the average accuracy of endoscopists (0.847 vs 0.777) was superior to that of the AI system. However, in detecting high-grade dysplasia, the AI's detection rate (80%) was significantly higher than that of the endoscopists (29.1%), and its diagnostic speed was much faster[48]. This suggests that while AI holds advantages in specific scenarios, its overall performance and generalizability still require improvement. Although no similar comparative studies specifically for GPLs exist yet, the limitations of AI revealed in this study are objective and widely prevalent.

Concurrently, there is a lack of robust clinical validation. The majority of current studies feature a retrospective design, which carries inherent limitations. Training data often originate from single centers, posing risks of selection bias and overfitting. The robustness and generalizability (i.e., performance across different devices and populations) of these models in diverse clinical environments await confirmation through larger-scale, prospective, multicenter randomized controlled trials (RCTs)[9]. A 2025 systematic review of 27 RCTs on AI in gastrointestinal oncology found that only 3 (11%) focused on GC[49]. Consequently, the feasibility, effectiveness, safety, and actual impact on long-term patient outcomes of AI systems within real-world clinical workflows necessitate validation through higher-level evidence[34,35]. Furthermore, the specific clinical scenarios where AI provides maximum decision-making support remain unclear. For instance, studies comparing the diagnostic accuracy of AI systems against endoscopists of varying experience levels have yielded inconsistent conclusions[48].

From the perspective of data and algorithms, significant challenges exist. The AI algorithms (e.g., CNN, support vector machine), imaging techniques (e.g., white light, narrow-band imaging), and study designs employed across different research exhibit high heterogeneity. This diversity, coupled with a lack of unified standards, makes it difficult to compare the merits of different models or identify the optimal combination of algorithms and technologies[50]. Moreover, high-quality, standardized training datasets for AI remain scarce[1]. The ratio of positive to negative samples in the training set influences AI diagnostic performance, with a 2022 study suggesting an optimal ratio between 1:1 and 1:2[51]. The quality, quantity, and diversity of the training dataset directly impact model performance. The datasets used in different studies vary greatly in image quality, annotation standards, and data scale. The absence of public, standardized large-scale endoscopic image databases (e.g., “EndoNet”) inevitably introduces selection bias. This not only limits the reproducibility and generalizability of results[9] but also hinders fair comparison between different algorithms and iterative model optimization[13]. Additionally, existing models are primarily trained on typical lesions, potentially limiting their ability to identify rare, morphologically atypical, or early minute lesions, thus posing a risk of missed diagnosis[41].

Another widely acknowledged critical limitation is that many AI models, particularly deep learning models, function as “black boxes”. Their decision-making process lacks transparency and interpretability, which can hinder clinicians' understanding, trust, and acceptance of the diagnostic rationale. This also complicates error troubleshooting and model optimization[1]. Although techniques like heatmaps can visualize the regions of attention for the AI[52], its underlying decision logic remains insufficiently transparent. This lack of clarity is certain to pose an obstacle to clinical application and widespread adoption[47]. Therefore, developing explainable AI (XAI) systems is a crucial future direction[43].

Finally, the regulatory framework remains underdeveloped. The development of unified data standards, reliable performance evaluation systems, and comprehensive regulatory and ethical frameworks is pivotal for the clinical translation of AI[13]. Among the most contentious issues is the undefined collaborative workflow between AI and human physicians. The specific role AI systems should play in clinical practice (e.g., primary screening tool, second opinion, or real-time monitoring) and how to optimize human-computer interaction to maximize diagnostic efficiency and accuracy require further exploration[48]. Table 3 outlines the major limitations and challenges associated with the clinical integration of AI for diagnosing GPLs.

Table 3 Major limitations and challenges in the clinical integration of artificial intelligence for diagnosing gastric precancerous lesions.

Category	Specific challenge/issue	Supporting evidence/explanation	Ref.
Generalizability and robustness	Performance degradation across centers	Most studies are single-center, leading to models that may not perform well on data from different hospitals, equipment, or populations	[47]
		A 2023 external validation study in Singapore found endoscopists’ average accuracy was superior to the AI system, highlighting generalizability issues	[48]
Clinical validation	Lack of high-level evidence	The majority of studies are retrospective with inherent limitations (selection bias, overfitting). Large-scale, prospective multicenter RCTs are needed	[9]
	Scarcity of RCTs in gastric cancer	A 2025 systematic review found only 11% of GI oncology AI RCTs focused on gastric cancer	[49]
	Unclear impact on patient outcomes	The feasibility, effectiveness, safety, and long-term impact on patient outcomes require validation	[34,35]
	Inconsistent conclusions on utility	Studies comparing AI against endoscopists of varying experience levels have yielded inconsistent results	[48]
Data and algorithms	High heterogeneity	High heterogeneity in algorithms, imaging techniques, and study designs makes comparing models difficult	[50]
	Lack of standardized, public datasets	Absence of public, standardized large-scale databases (e.g., “EndoNet”) introduces selection bias, limits reproducibility, and hinders fair comparison	[1,9,13]
	Suboptimal training data ratios	The positive-to-negative sample ratio in training sets influences performance, with a suggested optimal ratio between 1:1 and 1:2	[51]
	Potential for missed diagnosis	Models trained primarily on typical lesions may miss rare, atypical, or minute early lesions	[41]
Interpretability and trust	“Black box” problem	The decision-making process of many AI models lacks transparency, hindering clinician trust, acceptance, and error troubleshooting	[1]
		Although techniques like heatmaps can help, the underlying logic remains insufficiently transparent, posing an obstacle to adoption	[47,52]
Regulatory and workflow	Underdeveloped regulatory framework	Lack of unified data standards, reliable evaluation systems, and comprehensive regulatory/ethical frameworks	[13]
	Undefined human-AI collaboration	The specific role of AI (e.g., primary screening, second opinion) and how to optimize human-computer interaction require further exploration	[48]

AI: Artificial intelligence; RCT: Randomized controlled trial; GI: Gastrointestinal.

Advancing AI for diagnosing GPLs: Future research directions

Given the existing limitations, the potential iterative directions and research recommendations for AI application technology in GPLs have become well-defined.

Foremost is the initiation of large-scale, multicenter prospective studies. As with all medical research, AI-assisted diagnosis of GPLs requires high-level evidence. Large-scale, multicenter, prospective RCTs should be designed and implemented to evaluate the effectiveness, safety, generalizability, cost-effectiveness, and actual impact on clinical workflows and patient outcomes of AI-assisted diagnostic systems in real-world clinical settings. This is essential for validating their clinical value and providing high-level evidence to support the clinical application of AI technology[34,49]. Concurrently, validating the clinical utility of AI models in risk stratification, early diagnosis, and treatment response prediction for GPLs is crucial to bridge the gap between basic research and clinical application[53] and to strengthen translational research.

Simultaneously, the establishment of standardized datasets and evaluation platforms must be prioritized. This specifically involves encouraging the creation of public, standardized, high-quality large-scale gastric image databases containing images from diverse regions, different endoscopic equipment, and various pathological types. Efforts should be made to promote the development of international, multicenter, high-quality annotated endoscopic image and video databases, akin to “EndoNet”, and make them accessible to the research community. This will enhance research transparency and reproducibility, facilitating the standardized evaluation, fair comparison, and continuous improvement of algorithms[1]. Building upon standardization, targeted research addressing specific clinical challenges is necessary. Dedicated studies should be designed to evaluate and enhance AI’s capability to detect atypical, minute, or rare precancerous lesions, for instance, by employing augmented datasets enriched with more such cases for model training to reduce the miss rate. Guideline documents emphasize the need to establish rigorous evaluation frameworks, ethical guidelines, and regulatory policies to ensure model safety and reproducibility, and to validate their clinical translation potential[54,55]. Furthermore, academia and industry should collaborate to develop standardized guidelines for data acquisition, processing, and AI model validation to improve the reproducibility and reliability of research[53,56].

Additionally, head-to-head comparative studies are needed to explore optimal human-AI collaboration models. Research directly comparing different AI algorithms (e.g., CNNs with different architectures) or different imaging technology combinations (e.g., white light vs narrow-band imaging) should be conducted to identify the optimal AI technical solution for specific clinical scenarios (e.g., early cancer screening, invasion depth assessment)[20]. Meanwhile, through clinical simulation and prospective studies, the diagnostic performance of AI systems should be directly compared with that of endoscopists of varying experience levels, particularly regarding their ability to identify occult and early lesions. This will help explore different application modes for AI, such as a real-time assistance tool, a second reader, or a quality control monitor, thereby clarifying AI's optimal clinical positioning (e.g., as a primary screening tool or an expert-assistive device)[48]. Optimizing the human-computer interaction interface and workflow is essential for maximizing diagnostic efficiency and accuracy.

Of particular concern to many clinicians is the urgent need to develop XAI models. Future research should prioritize creating interpretable AI models. Techniques such as heatmaps, feature attribution, saliency maps, attention mechanisms, feature visualization, or the integration of multimodal data can be employed to visualize the diagnostic basis of models, reveal the biological rationale behind their decisions[57], and make the decision-making process transparent to clinicians, thereby enhancing trust and efficiency in human-AI collaboration[47]. These models should be able to explicitly indicate the key image features or data points upon which they base their diagnostic or predictive judgments, strengthening clinicians’ understanding and trust in the AI decision-making process. Within the modern medical system built on an evidence-based foundation, this aspect is potentially most critical; the reliability of evidence will significantly impact AI’s application prospects and its acceptance within the medical community.

Furthermore, exploring multimodal data fusion represents a more advanced step. Future research is recommended to integrate standardized, high-quality multimodal data-such as endoscopic images, pathological slides, genomic information, and clinical data-to construct more comprehensive and precise fused AI models for diagnosis and prognosis prediction. This enhances model performance and is used to train and validate more robust AI models[1], aiming to build more thorough and accurate risk stratification and prognostic prediction models for GC, ultimately enabling deeper levels of personalized medicine[47]. Compared to the previously mentioned research directions, this represents a more advanced frontier. It may not carry the same immediate urgency nor pose a fundamental threat to the prospects of AI in medical applications. However, after addressing the aforementioned limitations, it represents a vast and promising field for the limitless future development of AI in the domain of GPLs. Table 4 proposes future research directions for AI in the diagnosis of GPLs.

Table 4 Proposed future research directions for artificial intelligence in diagnosing gastric precancerous lesions.

Research direction	Specific goals/actions	Expected outcomes/rationale	Ref.
High-quality clinical trials	Conduct large-scale, multicenter, prospective RCTs	Evaluate real-world effectiveness, safety, generalizability, cost-effectiveness, and impact on patient outcomes to provide high-level evidence for clinical adoption	[34,49]
	Validate clinical utility in risk stratification and prediction	Bridge the gap between basic research and clinical application, strengthening translational research	[53]
Data standardization and infrastructure	Create public, standardized, large-scale image databases	Contain diverse images from different regions, equipment, and pathologies. Enhance research transparency, reproducibility, and facilitate fair algorithm comparison and improvement	[1]
	Develop international, multicenter, annotated databases (e.g., “EndoNet”)	Make high-quality data accessible to the research community to overcome a major current limitation
	Establish guidelines for data acquisition and model validation	Improve the reproducibility and reliability of AI research through academia-industry collaboration	[53,56]
Targeted algorithm development	Enhance detection of atypical/minute lesions	Employ augmented datasets enriched with rare cases for training to reduce miss rates and improve robustness
	Establish rigorous evaluation frameworks and ethical guidelines	Ensure model safety, reproducibility, and validate clinical translation potential	[54,55]
Comparative and optimization studies	Head-to-head comparison of AI algorithms and imaging techniques	Identify the optimal AI technical solution for specific clinical scenarios (e.g., screening vs depth assessment)	[20]
	Explore optimal human-AI collaboration models	Compare AI performance against endoscopists of varying experience; explore modes like real-time assistance, second reader, or quality control monitor to clarify AI’s best clinical role	[48]
	Optimize human-computer interaction interface and workflow	Maximize diagnostic efficiency and accuracy in clinical practice
Explainable AI	Develop interpretable AI models using heatmaps, attention mechanisms, etc.	Visualize the diagnostic basis of models, make the decision-making process transparent to clinicians, and enhance trust and efficiency in human-AI collaboration	[43,47,57]
Multimodal data fusion	Integrate endoscopic images, pathology, genomics, and clinical data	Build more comprehensive, robust, and accurate fused AI models for diagnosis, risk stratification, and prognosis prediction, enabling personalized medicine	[14,47]

AI: Artificial intelligence; RCT: Randomized controlled trial.

APPLICATION OF ORGANOID TECHNOLOGY IN CANCER RESEARCH

In contrast to AI’s focus on diagnosis, organoid technology primarily provides advanced in vitro models for mechanistic studies of GC and its precancerous lesions. Traditional two-dimensional (2D) cell cultures and animal models are insufficient for simulating the complex pathophysiological processes of human GPLs. For instance, animal models may fail to fully recapitulate the human process of IM or skip critical precancerous stages[6]; meanwhile, 2D cell lines often lose their resemblance to the original tumor due to selective pressure from genetic variations[16]. Furthermore, the development and progression of GC are closely associated with a complex TME, encompassing immune cells, stromal cells, and the intratumoral microbiota[17]. Traditional 2D cell cultures and animal models struggle to fully replicate these intricate interactions.

To overcome these limitations, researchers urgently need in vitro models that can more faithfully simulate the disease process. Organoid technology has emerged to meet this need. Utilizing 3D stem cell culture techniques, it enables the construction of miniaturized organs in vitro that retain the genetic, phenotypic, and behavioral characteristics of the original tissue[16,17,58]. Patient-derived organoids (PDOs) can recapitulate many features of their source tissue, including organizational structure, cell subpopulations, and the capacity to respond to specific patient “disease information”. This offers unprecedented opportunities for basic research and translational medicine, particularly in the fields of personalized therapy and drug screening[58,59]. In recent years, GC organoids (GCOs) have become an important platform for studying GC heterogeneity, the TME, immune escape, and drug sensitivity[17,60,61].

Organoids as a potential ideal platform for researching the pathogenesis of GPLs and for drug screening

A 2025 review indicated that GCOs can preserve the genetic, phenotypic, and behavioral characteristics of the source tumor tissue, making them ideal models for studying disease heterogeneity and the dynamic TME[17]. By constructing co-culture organoid systems containing immune cells, researchers can deeply explore how tumor cells evade surveillance and attack by the immune system-for example, by analyzing the relationship between the distribution of tumor-infiltrating lymphocytes and the response to immune checkpoint inhibitor therapy[38]. This provides a potential means to understand the immune escape mechanisms involved in the progression from precancerous lesions to invasive carcinoma.

Organoid technology provides a powerful in vitro model for gaining a deeper understanding of the molecular mechanisms of GPLs, thereby facilitating the discovery and validation of diagnostic biomarkers. A 2024 study investigated the role of OLFM4 in incomplete IM (IIM), a high-risk precursor lesion of GC. This research utilized organoid models established from IIM tissue biopsies. The results showed that OLFM4 was overexpressed in IIM organoids. Further investigation using this model revealed that OLFM4 binds to MYH9, accelerates the ubiquitination of GSK3β, leads to increased β-catenin levels, and consequently promotes the proliferation and invasion of GPL cells through the Wnt signaling pathway. This organoid-based study not only uncovered a key cellular signaling pathway in the progression of IM but also confirmed OLFM4 as a novel biomarker for IIM, whose diagnostic accuracy surpasses that of traditional markers CDX2 and MUC2, offering an auxiliary means for EGC screening[62].

Furthermore, multiple review articles emphasize that organoids' ability to retain the genetic and phenotypic characteristics of the original tissue makes them ideal models for studying tumorigenesis, progression, and treatment response[16,58]. Some studies mention that GCOs can preserve their genetic, phenotypic, and behavioral features, demonstrating significant potential in simulating the TME, exploring immune escape mechanisms, and predicting individualized treatment responses[17]. Through in-depth analysis of organoids derived from GPL tissues, researchers can explore their unique molecular signatures and discover potential biomarkers associated with disease progression. This suggests that establishing PDO models from patients with GPLs enables detailed study of the molecular mechanisms driving the evolution from precancerous lesions to GC and provides an ideal in vitro testing platform for developing preventive interventions. Additionally, this technology is also used to study the role of the intratumoral microbiota [e.g., Helicobacter pylori (H. pylori)] in the initiation and development of GC.

As an advanced 3D in vitro model, organoids can highly simulate the structure and function of in vivo tissues, preserving the genetic, histological, and pathophysiological characteristics of the original tumor[18]. This property grants them an irreplaceable advantage in drug sensitivity testing and new drug development for oncology. According to the “Chinese Expert Consensus on Standardized Establishment and Clinical Application of Endometrial Carcinoma Organoids (2025 Edition)”, organoid models have been applied in various malignant tumors, including esophageal, gastric, colorectal, and bladder cancers. They have been proven to reflect tumor heterogeneity, and their drug responses correlate with those of the corresponding patients, providing a novel solution for achieving precision medicine[18]. Organoid models can not only serve as high-throughput drug screening platforms to discover new therapeutic drugs for specific patients or particular molecular subtypes[17], but other studies have also shown that GCOs can be used to predict patient responses to chemotherapy, radiotherapy, targeted therapy, and immunotherapy, making individualized treatment a possibility[15]. For example, conducting drug sensitivity tests on organoids can help clinicians select the most effective treatment regimen for specific patients[63].

Limitations and future directions of organoid models in GPL research

Although organoid technology shows promising prospects and demonstrates significant potential in GPL research, related investigations remain in their nascent stages. Current studies predominantly focus on advanced GC, while research dedicated to constructing organoid models for precancerous lesions such as atrophic gastritis and IM, and elucidating their evolutionary mechanisms, is relatively limited[17].

Insufficient biological complexity and fidelity represent a core limitation of current models. Although organoids surpass traditional models in tumor simulation, existing organoid systems still face constraints in fully recapitulating vascular systems, innervation, and interactions with systemic physiological processes[17]. Multiple review articles emphasize that current organoid models consist primarily of epithelial cells and lack critical components of the complex TME, including immune cells, stromal cells, and intratumoral microbiota[16-18,54,64,65]. Given that GC development is closely associated with H. pylori infection and inflammatory responses[6,66], pure epithelial organoids cannot fully replicate these essential interactions. Consequently, developing co-culture systems incorporating multiple cell types to better simulate the in vivo microenvironment represents a crucial direction for future research[17].

Furthermore, standardization remains a significant technical challenge in organoid culture methodologies, analytical techniques, and data interpretation[17,67]. A 2024 review emphasizes the necessity of establishing well-characterized organoid models encompassing all GC subtypes to achieve genuine precision medicine[60]. For instance, a 2024 commentary on the aforementioned organoid-based biomarker discovery study provided methodological recommendations, suggesting that while organoids represent advanced models, more established systems like air-liquid interface (ALI) models should be considered for biomarker research, advocating for more rigorous and standardized approaches to ensure clinical applicability and reliability of research findings[67]. Beyond standardization deficiencies, organoid research also confronts challenges in scalability, reproducibility, cost-effectiveness, and time efficiency[68-70]. The “Recommended Guidelines for Developing, Qualifying, and Implementing Complex In Vitro Models (CIVMs) for Drug Discovery” highlight the gap between model development and qualification, noting that absent standardized qualification processes undermine confidence in models' physiological relevance[55]. The relatively long culture cycle, variable success rates, batch-to-batch variations, and high costs of organoids collectively limit their large-scale clinical application and regulatory acceptance[54,55,68-70].

Although organoids preserve certain aspects of tissue heterogeneity, whether they fully represent all characteristics of the original lesional tissue requires further validation. Organoid technology currently suffers from inconsistent culture success rates and extended culture cycles[15]. Additionally, the lack of unified standards for organoid culture, establishment, and analytical protocols may compromise the reproducibility and reliability of research outcomes[17,67], raising questions about the representativeness of organoid models. Moreover, despite being considered advanced models, there is a notable absence of head-to-head comparative studies pitting organoids against more established models (e.g., animal models, ALI models) in simulating GPLs, which would clarify their unique advantages and optimal applications[6,67].

Another significant consideration is the limited evidence supporting the clinical translation of organoid technology[17,67]. Currently, organoids are primarily utilized in basic research for mechanistic exploration and biomarker discovery[17,62]. Direct evidence for their application in clinical diagnostics (e.g., predicting lesion progression risk through organoid models) remains scarce, and the technology remains distant from direct clinical implementation[61]. A 2025 review noted that although GCOs constitute a highly promising platform, the technology requires further advancement before clinical application and necessitates ongoing technical improvements to facilitate its role in basic and translational oncology research[61]. Table 5 details the major limitations and challenges of organoid models in gastric GPL research.

Table 5 Major limitations and challenges of organoid models in gastric precancerous lesion research.

Category	Specific challenge/issue	Supporting evidence/explanation	Ref.
Research focus	Relative scarcity of precancerous lesion models	Current studies predominantly focus on advanced gastric cancers, with limited research on constructing organoid models for atrophic gastritis and intestinal metaplasia	[17]
Model fidelity and complexity	Lack of tumor microenvironment (TME) components	Existing models primarily consist of epithelial cells and lack critical TME components (immune cells, stromal cells, intratumoral microbiota), unable to fully replicate essential interactions (e.g., with H. pylori)	[16-18,54,64,65]
	Inability to recapitulate systemic physiology	Constraints in fully recapitulating vascular systems, innervation, and interactions with systemic physiological processes	[17]
Standardization and reproducibility	Lack of standardized protocols	Significant challenges exist in organoid culture methodologies, analytical techniques, and data interpretation, compromising reproducibility and reliability	[17,67]
	Model qualification gap	Absent standardized qualification processes undermine confidence in models' physiological relevance	[55]
Scalability and practicality	Challenges in scalability and cost	Limitations in scalability, reproducibility, cost-effectiveness, and time efficiency. Relatively long culture cycle, variable success rates, batch-to-batch variations, and high costs limit large-scale application	[54,55,68-70]
Model validation and comparison	Unclear representativeness	Whether organoids fully represent all characteristics of the original lesional tissue requires further validation. Inconsistent culture success rates and extended cycles are current drawbacks	[15]
	Lack of comparative studies	Notable absence of head-to-head studies comparing organoids against more established models (e.g., animal models, ALI models) to clarify their unique advantages and optimal applications	[6,67]
Clinical translation	Limited direct clinical evidence	Organoids are primarily used in basic research. Direct evidence for application in clinical diagnostics (e.g., predicting progression risk) remains scarce, and the technology remains distant from direct clinical implementation	[17,61,62,67]

ALI: Air-liquid interface; H. pylori: Helicobacter pylori.

The enumeration of these limitations leads to proposed future research directions for organoids in the field of GPLs.

First and most fundamental is the development and validation of GPL organoid models. Efforts should focus on successfully establishing organoid models from tissues of patients with GPLs (e.g., IM) and validating their ability to simulate the process of malignant transformation in vitro. This will enable the study of key molecular events and driver genes[17].

The most pressing need is the development of more complex co-culture organoid models. To address the insufficient fidelity of current models, future research should aim to create co-culture organoid or “organoid-on-a-chip” systems that incorporate vascular networks, immune cells (e.g., T cells, macrophages), and stromal cells (e.g., fibroblasts). Integrating elements such as H. pylori into GPL organoid culture systems would allow the construction of co-culture models that more authentically simulate the in vivo microenvironment. This would facilitate the study of interactions between different components and their impact on lesion progression. For instance, microfluidic technology could be employed to create “tumor-on-a-chip” models that more accurately mimic the complex in vivo TME, providing more reliable platforms for investigating immune escape mechanisms and developing immune prevention strategies. Another instance would be, 3D bioprinting technology could be utilized to construct more precisely structured TMEs that better simulate the in vivo response to drugs, particularly immunotherapies[54,64,65].

Equally important is the establishment of standardized organoid biobanks, standard operating procedures (SOPs), and quality control systems. The development of GPL organoid biobanks containing detailed clinical and pathological information should be promoted. Concurrently, standardized SOPs should be formulated to enhance the consistency and comparability of research. Through multicenter collaborations and industry alliances, joint efforts should be made to establish standardized operating procedures for organoid culture qualification, functional analysis, drug screening, data acquisition, and analysis. This includes implementing strict quality control measures for cell sources, culture medium components, 3D scaffold materials, and key performance indicators (e.g., morphology, gene expression profiles) to improve the reproducibility and comparability of research results[54,55,68].

To overcome existing limitations and address challenges related to scalability, reproducibility, cost-effectiveness, and time efficiency, researchers are committed to developing high-throughput screening (HTS) platforms. These platforms aim to enhance the efficiency of organoid culture and drug testing through automation and microfluidic technologies[70-72]. Although automation and microfluidics offer improvements, achieving an optimal balance between low cost, high throughput, and maintained biological complexity remains challenging, particularly when processing large-scale patient samples[55,70,72]. Therefore, further optimization of HTS platforms and cost reduction are necessary. Continued exploration of technologies such as microfluidics, acoustic manipulation, automated liquid handling, and high-content imaging is essential to increase the throughput and automation level of organoid screening. Simultaneously, research into low-cost alternative materials, such as synthetic hydrogels to replace Matrigel, should be conducted to reduce the economic burden of large-scale screening and promote clinical translation[70,72].

Building upon these foundations, conducting multi-model comparative studies is recommended. Research should be designed to directly compare the advantages and limitations of organoid models, genetically engineered mouse models, chemically induced animal models, and ALI models in simulating the characteristics of each stage of the “Correa cascade”. This would help determine the most suitable model for different research objectives and clarify the applicable scenarios for organoid technology.

Furthermore, exploring the application of organoids in predictive diagnostics represents an ideal research direction bridging basic and clinical science. Prospective studies should be conducted using organoid models established from tissues of patients with different risk grades (e.g., low-grade vs high-grade intraepithelial neoplasia). By integrating high-throughput technologies such as single-cell sequencing, these studies could identify molecular features predictive of lesion progression or malignant transformation, providing a foundation for developing organoid-based personalized risk assessment tools. Table 6 lists proposed future research directions for organoid models in GPL research.

Table 6 Proposed future research directions for organoid models in gastric precancerous lesion research.

Research direction	Specific goals/actions	Expected outcomes/rationale	Ref.
Model development	Establish precancerous lesion organoid biobanks	Develop organoid models from patient tissues (e.g., with IM) and validate their ability to simulate malignant transformation in vitro, enabling study of key molecular events and driver genes	[17]
Enhanced complexity (co-culture)	Develop complex co-culture systems	Create co-culture organoid or “organoid-on-a-chip” systems incorporating vascular networks, immune cells (T cells, macrophages), stromal cells (fibroblasts), and H. pylori	[54,64,65]
		Utilize microfluidic technology for “tumor-on-a-chip” models to accurately mimic the complex TME for studying immune escape and prevention
		Employ 3D bioprinting to construct precise TMEs that better simulate in vivo responses to drugs, particularly immunotherapies
Standardization and biobanking	Establish SOPs and quality control systems	Promote the development of biobanks with detailed clinical/pathological information. Formulate standardized SOPs for culture, qualification, functional analysis, and data handling to enhance consistency and comparability	[54,55,68]
High-throughput screening (HTS)	Develop automated HTS platforms	Enhance the efficiency of organoid culture and drug testing through automation, microfluidics, acoustic manipulation, and high-content imaging	[70-72]
	Reduce costs	Explore low-cost alternative materials (e.g., synthetic hydrogels) to replace Matrigel, reducing the economic burden for large-scale screening and promoting translation	[70,72]
Comparative studies	Conduct multi-model comparison studies	Directly compare organoids vs GEMMs, chemical animal models, and ALI models in simulating the “Correa cascade” to determine the best model for specific research objectives and clarify organoid applications	[6,67]
Predictive diagnostics	Explore application in risk stratification	Use organoids from patients with different risk grades (e.g., LGIN vs HGIN) integrated with scRNA-seq to identify molecular features predictive of progression, enabling personalized risk assessment

ALI: Air-liquid interface; GEMM: Genetically engineered mouse model; HGIN: High-grade intraepithelial neoplasia; IM: Intestinal metaplasia; LGIN: Low-grade intraepithelial neoplasia; scRNA-seq: Single-cell RNA sequencing; SOP: Standard operating procedure; TME: Tumor microenvironment; H. pylori: Helicobacter pylori.

SYNERGISTIC APPLICATION PROSPECTS OF AI AND ORGANOID TECHNOLOGY

Although there are no existing precedents for the synergistic application of AI and organoid technology in GPL research, the significant potential of their integration can be clearly outlined based on the respective strengths of each technology. Existing research has systematically described how AI can utilize multi-omics data, spatial omics technologies, and organoid models to analyze the molecular mechanisms of diseases (especially tumors) and identify key regulatory networks and biomarkers. This provides a solid foundation for conceptualizing and designing potential logical pathways for their synergistic application.

Synergistic applications of AI and organoid technology

The convergence of AI and organoid technology offers a powerful synergistic framework that bridges basic research and clinical translation. This integration can be conceptualized across three interconnected domains: The application of AI in deciphering complex molecular data, its role in high-throughput organoid phenotyping, and the ultimate fusion of multimodal data to develop predictive models for precision medicine. The synergistic pipeline connecting patient data, organoid models, and AI prediction is depicted in Figure 1.

Figure 1 The integrative paradigm of artificial intelligence and organoid technology for gastric precancerous lesion research. A: A detailed schematic diagram of the artificial intelligence (AI)-organoid synergy pipeline. The framework begins with the collection of clinical data (endoscopic/pathological images and genomics) and tissue biopsies from patients. Biopsies are used to establish patient-derived organoids, which are subjected to high-throughput dynamic imaging and drug screening. The multimodal AI integration and prediction model (central core) fuses all clinical, imaging, and drug response data. This integrated analysis outputs personalized risk stratification and preventive intervention strategies, enabling precision prevention. Arrows indicate the direction of data and sample flow; B: A simplified closed-loop flowchart summarizing the core logic of the AI-organoid integration. The model depicts the cyclical process where clinical data and biopsies from the patient inform the generation and analysis of organoid models. The resulting data is processed by the multimodal AI prediction model to generate personalized outputs that guide clinical decision-making, ultimately feeding back to improve patient care. AI: Artificial intelligence; 3D: Three-dimensional.

AI for multi-omics and spatial multi-omics analysis: Multi-omics analysis has emerged as a cornerstone of life sciences research, integrating datasets from genomics, epigenomics, transcriptomics, proteomics, and metabolomics to systematically unravel the molecular mechanisms of disease pathogenesis[73]. In oncology, this approach provides unprecedented insights into tumor heterogeneity and facilitates the discovery of novel biomarkers[74]. The recent advent of spatial multi-omics technologies further enables the examination of gene and protein expression within their native tissue context, permitting in-depth dissection of the TME, cellular heterogeneity, and intercellular communication networks[75].

A significant challenge, however, lies in the inherent characteristics of multi-omics data-high dimensionality, noise, and heterogeneity-which complicate the extraction of biologically meaningful insights[73,76]. AI, particularly machine learning and deep learning, offers powerful solutions to these challenges by leveraging its superior data processing and pattern recognition capabilities[77,78]. It can integrate and analyze diverse data streams, automatically extract complex features, and demonstrate considerable potential in cancer screening, diagnosis, molecular subtyping, and prognosis prediction[53,73,78].

AI is pivotal for integrating and analyzing these massive, high-dimensional datasets to identify key molecular mechanisms and biomarkers. By synthesizing multi-layered information, AI models can reconstruct intricate biological networks and elucidate the molecular underpinnings of disease[73,77]. For instance, AI algorithms effectively reduce the dimensionality of multi-omics data and analyze complementary multimodal data streams, driving progress in precision oncology[73]. The synergy between AI and network biology also advances the understanding of complex phenotypes, such as tumor multidrug resistance[77]. This methodology is supported by expert consensuses, which recognize that deep learning-based multi-omics models can predict cancer survival rates and identify molecular features associated with aggressive subtypes[79].

In the spatial domain, AI is indispensable for processing and interpreting complex spatial multi-omics data. AI-based multimodal models can uncover the molecular interactions governing cellular behavior and tissue dynamics, thereby predicting critical disease stages and informing therapeutic strategies[75]. Compared to conventional histopathology, highly multiplexed spatial proteomics data yield deeper mechanistic insights, with deep learning serving as a vital tool for deciphering this information[80].

AI for organoid image analysis and phenotyping: As advanced in vitro 3D culture models, organoids emulate the structure and function of real organs[81,82], providing invaluable platforms for disease modeling and drug screening[83]. A major challenge, however, is the complexity of the organoid culture process, which generates vast amounts of data that are often inefficiently analyzed and prone to error[84].

The cultivation and analysis of organoids produce extensive image data (e.g., from bright-field and fluorescence microscopy). AI algorithms can be trained to automatically identify and quantify morphological features (e.g., size, shape, complexity), monitor growth dynamics, and assess responses to pharmacological interventions. This automated, high-throughput image analysis significantly enhances the efficiency and objectivity of organoid-based screening, overcoming the time-consuming and subjective limitations of manual evaluation.

Multiple reviews highlight AI’s capacity to derive insights and predictions from diverse organoid data sources, including microscopic images, transcriptomics, and proteomics[83,84]. Its applications encompass label-free identification, 3D image reconstruction, quality control, and streamlined multi-omics data analysis, enabling precise preclinical assessment[83,84]. Expert consensuses actively advocate for the use of AI in organoid research, such as optimizing culture conditions, material design, and printing parameters to achieve precise structural control of organoids[85]. The immense and complex datasets generated by advanced analytical techniques (e.g., proteomics, metabolomics, single-cell RNA sequencing, spatial transcriptomics) represent an area where AI can demonstrate particular utility[85].

Although direct applications in GPLs are still emerging, the underlying technical framework is well-established. Combining long-term live-cell imaging of organoids with AI-based analysis allows for the dynamic tracking of cellular behaviors (e.g., proliferation, differentiation, apoptosis) and their correlation with multi-omics data at specific timepoints. This approach can elucidate the dynamic regulatory processes of key molecular events under specific stimuli, substantially advancing mechanistic research.

The integrative paradigm: Multi-modal data fusion from patients and organoids: Multimodal data integration constitutes the core of this synergistic application, with AI serving as the unifying computational engine. AI can amalgamate multidimensional patient data-including endoscopic images, pathological slides, and genomic profiles-with drug response data acquired from PDO models[4,73]. Constructing multimodal AI models enables more accurate prediction of the progression risk of precancerous lesions.

For example, an AI model could integrate the endoscopic image features of a patient with the response of their corresponding organoids to specific chemopreventive agents. This integrated model could then predict the patient's future probability of developing GC, thereby providing a more robust foundation for clinical decision-making. Furthermore, AI can fuse multi-omics data (e.g., genomics, transcriptomics) with imaging data derived from organoids to build multimodal predictive models for forecasting the progression risk of precancerous lesions or the response to specific preventive interventions[3].

The integration of AI technology with organoid research not only optimizes organoid establishment and culture conditions but also facilitates efficient analysis of their microscopic images and multi-omics data, thereby accelerating both research progress and clinical translation[83,84]. This integrative paradigm, which merges deep clinical phenotyping with high-fidelity in vitro modeling, represents a transformative frontier in shifting the management of GPLs from detection towards mechanistic understanding and precise prevention.

Prospects of combining organoids and AI in drug screening and clinical applications

Compared to traditional 2D cell cultures and animal models, organoids offer significant advantages in reflecting human physiology, genetic variation, and disease mechanisms. They can better recapitulate the histological and molecular characteristics of donor tissues and simulate both intra-tumoral and inter-tumoral heterogeneity[18,68]. Consequently, organoids are widely regarded as a highly promising platform for disease modeling, drug screening, and regenerative medicine[54,81,86].

In the field of drug discovery, high research and development costs and the high failure rate of clinical trials are long-standing challenges[55]. The limited predictive capacity of traditional preclinical models (e.g., 2D cell lines) for clinical efficacy leads to the failure of many candidate drugs in later development stages due to lack of effectiveness or safety concerns[55]. PDOs, which retain the histological and genetic features of the primary tumor, can serve as a platform for predicting individualized therapeutic efficacy, offering new possibilities for precision medicine[18,86-88]. Currently, organoid models have demonstrated significant clinical value in drug sensitivity testing across various solid tumors, such as colorectal, ovarian, breast, pancreatic, and endometrial cancers[18,72,87-89].

Firstly, integrating organoid HTS with AI models can enhance the accuracy of drug response prediction. Combining organoid HTS platforms with AI is a key strategy for accelerating drug discovery and achieving personalized medicine. Multiple studies and guidelines have emphasized the potential and necessity of this integrated approach[55,69,81,86,90]. On one hand, advances in automation and microfluidics have facilitated the establishment of organoid HTS platforms. These platforms enable large-scale, standardized, and reproducible drug testing[70,72]. On the other hand, AI and machine learning models provide powerful tools for analyzing the massive data generated by HTS. Deep learning models are used for automated segmentation and analysis of organoid images, significantly improving the efficiency and accuracy of drug screening evaluations. Furthermore, AI models can be directly applied to predict drug responses. A 2025 study developed a predictive model named PharmaFormer, based on the Transformer architecture. This model was first pre-trained on extensive 2D cell line data and then fine-tuned using limited organoid pharmacogenomic data. The results showed that this transfer learning strategy significantly improved the accuracy of predicting clinical drug responses, demonstrating that combining AI models with organoid data can accelerate the advancement of precision medicine[69]. Similarly, a 2024 study developed the VirtuDockDL platform, which uses graph neural networks to predict compound efficacy, achieving 99% accuracy on the HER2 dataset, far surpassing other methods[91]. Collectively, this evidence indicates that integrating automated HTS platforms to generate high-quality data, coupled with advanced AI models for in-depth analysis and prediction, can establish an efficient and precise preclinical drug evaluation system[55,90].

Moreover, the combination of organoids and AI holds potential for specific tumor types and treatment strategies. By conducting high-throughput drug screening on a large number of PDOs and using AI to analyze their association with genomic and transcriptomic data, novel molecular markers associated with drug sensitivity or disease progression can be identified[92]. These newly discovered markers can not only be used for patient risk stratification but also serve as potential targets for developing therapeutic or preventive strategies. Platforms combining organoids and AI models have demonstrated substantial potential in the individualized treatment of various cancers, guiding decisions for chemotherapy, endocrine therapy, targeted therapy, and even immunotherapy[18,64,88]. Additionally, organoid scoring systems provide a quantitative tool for predicting clinical outcomes. A 2022 study published in Nature Communications developed an “organoid score” for 54 colorectal cancer patients. It found that a higher organoid score was significantly associated with lower tumor regression rates and shorter progression-free survival in patients receiving standard treatment, proving the value of this scoring system in predicting treatment response and disease progression[89]. These application examples fully illustrate that the integration of organoids and AI extends beyond “repurposing existing drugs” and can also accelerate the development and clinical translation of new targets and therapies[18].

Challenges, opportunities, and future directions for the integration of AI and organoids in the field of GPLs

The previous sections have provided a relatively detailed discussion of the individual limitations of AI and organoids. Here, we supplement this by addressing the challenges faced in their synergistic application.

First, there is a research gap in the integration of AI and organoid technology. Current evidence independently explores the application of either AI or organoids in GC (or oncology more broadly). Almost no studies combine these two technologies, and there is a complete lack of empirical data on the application of their integration specifically to the research of GPLs[1]. As of the completion of this review, in the field of GPLs, there is an absence of research combining AI technology (e.g., pathological image analysis) with organoid models. This precludes the use of AI for high-throughput, quantitative analysis of the dynamic changes in organoid models to reveal the evolutionary patterns of precancerous lesions[38].

Second, data analysis and model validation pose significant challenges. The performance of AI models is highly dependent on high-quality, large-scale training data, yet currently available public organoid pharmacogenomic data linked to clinical outcomes remain limited[69].

Concurrently, data standardization presents a challenge. Effective integration of AI and organoid research requires standardized processes for data acquisition, processing, and annotation, encompassing both imaging data and organoid culture/analysis data. Currently, unified standards in these areas are lacking[1,13,93].

Additionally, challenges exist in data integration and handling heterogeneity. Even with AI methods, integrating multi-omics data from different platforms and batches, or even multi-dimensional data from organoid studies, often presents technical difficulties. The heterogeneity, sparsity, and high dimensionality of the data are major obstacles that current computational methods must overcome[53,75,76].

The most uncertain issue currently is the translational gap from research to clinic. Despite significant progress in both AI and organoid research, the seamless integration of these findings into clinical workflows remains limited. Effectively translating laboratory discoveries into clinical applications that can improve patient outcomes is a key challenge currently faced[53,56,74].

Thus, moving forward, extensive groundwork must be addressed to advance the collaborative integration of both technologies.

Fundamentally, exploring the direct collaborative integration of AI and organoids is paramount. It is recommended to initiate innovative research applying AI image analysis technology to the dynamic monitoring of organoid models. For example, using AI for real-time, label-free quantitative analysis of organoid morphological changes, cell proliferation, and differentiation to high-throughput screen for factors or drugs influencing precancerous lesion progression.

Integrated prospective research projects can be launched. Design and conduct prospective, multicenter clinical studies to simultaneously collect endoscopic/pathological imaging data and tissue samples from cohorts of patients with GPLs. Use the tissue samples to establish PDOs models and perform drug screening or disease progression simulation. Ultimately, use AI models to integrate imaging features and organoid phenotypic data to validate the value of the synergistic application in risk prediction and intervention decision-making; or use AI to analyze their response to different chemopreventive drugs.

There is a need to construct multicenter, standardized databases. Establish a large-scale, multicenter database for GPLs containing standardized clinical information, endoscopic images, digital pathology images, multi-omics data, corresponding organoid culture and analysis data, and even associated drug response data and clinical outcome information[69,90]. This would provide a high-quality data foundation for developing and validating AI models with greater generalizability[1].

Furthermore, for specific clinical challenges, specialized translational studies should be designed based on established generic technologies to address unmet needs. For instance, the application discussed in this paper concerning the integration of AI and organoids in the field of GPLs. For example, CRISPR-Cas9 gene editing technology could be used to simulate relevant gene mutations in organoids, establishing disease models with abnormal folate metabolism pathways or high RAMP1 expression. AI-assisted high-throughput platforms could then be used to screen and optimize targeted drugs, validating their efficacy as personalized treatment strategies.

Advanced data integration algorithms need to be developed. More resources must be invested in developing advanced AI algorithms capable of effectively handling data heterogeneity, imputing data sparsity, and integrating multimodal data (e.g., imaging, spatial omics, single-cell omics) to build more comprehensive disease models[74,75], and to achieve better multidimensional integration with organoid co-culture systems. Future research should combine AI-driven dynamic imaging with multi-omics analysis to investigate complex cell-cell interactions.

Furthermore, an AI-driven high-throughput organoid analysis platform can be established. Develop a dedicated AI image analysis platform for organoid research. This platform should be capable of automatically performing dynamic monitoring and quantitative analysis of processes such as organoid growth, differentiation, and death, enabling high-throughput, automated interpretation of drug screening results, and accelerating the translation from basic research to clinical application. Table 7 presents the key challenges and future research directions for the integration of AI and organoids in GPL research.

Table 7 Challenges and future directions for the integration of artificial intelligence and organoids in gastric precancerous lesion research.

Category	Specific challenge/opportunity	Explanation/proposed Action	Expected outcome/goal	Ref.
Current challenges
Research gap	Lack of integrated studies	Almost no studies combine AI and organoid technologies. A complete lack of empirical data on their integrated application specifically to GPLs	N/A	[1]
Data dependency and validation	Limited training data	AI model performance depends on high-quality, large-scale data, yet public organoid pharmacogenomic data linked to clinical outcomes remain limited	N/A	[69]
Data standardization	Lack of unified standards	Effective integration requires standardized processes for data acquisition, processing, and annotation (imaging and organoid data). Currently, such standards are lacking	N/A	[1,13,73]
Data integration	Handling data heterogeneity	Integrating multi-omics and multi-dimensional data from different platforms/batches presents technical difficulties due to heterogeneity, sparsity, and high dimensionality	N/A	[53,76,77]
Clinical translation	Translational gap	Seamlessly integrating laboratory findings from AI and organoids into clinical workflows to improve patient outcomes remains a key challenge	N/A	[53,56,75]
Future directions
Direct integration	AI for organoid dynamic monitoring	Initiate research applying AI for real-time, label-free quantitative analysis of organoid morphological changes, proliferation, and differentiation	To high-throughput screen for factors or drugs influencing precancerous lesion progression	[38]
Prospective studies	Integrated clinical research projects	Design prospective, multicenter studies to collect imaging data and tissue samples. Build PDOs for screening/simulation, then use AI to integrate imaging and organoid data	To validate the value of synergy in risk prediction and intervention decision-making
Database construction	Multicenter standardized database	Establish a large-scale database containing clinical info, endo/pathology images, multi-omics, organoid culture/analysis, drug response, and outcome data	To provide a high-quality data foundation for developing and validating generalizable AI models	[1,69,91]
Targeted translational studies	Address specific unmet clinical needs	Use CRISPR-Cas9 in organoids to simulate mutations (e.g., abnormal folate metabolism, high RAMP1). Use AI-high-throughput platforms to screen/optimize targeted drugs	To validate efficacy for personalized treatment strategies
Algorithm development	Advanced data integration algorithms	Develop AI algorithms capable of handling data heterogeneity, imputing sparsity, and integrating multimodal data (imaging, spatial omics, scRNA-seq)	To build more comprehensive disease models and achieve better integration with organoid co-culture systems	[75,76]
Platform establishment	AI-driven HTS organoid analysis platform	Develop a dedicated AI image analysis platform for automated, quantitative dynamic monitoring of organoid growth, differentiation, and death	To enable high-throughput interpretation of drug screening results and accelerate translation from bench to bedside

AI: Artificial intelligence; HTS: High-throughput screening; PDOs: Patient-derived organoids; scRNA-seq: Single-cell RNA sequencing; GPL: Gastric precancerous lesion.

Of particular importance are the significant regulatory and ethical challenges confronting the clinical translation of AI-organoid integrated technologies, where regulatory hurdles include the current lack of standardized certification for drugs screened via AI-analyzed organoids, necessitating novel evaluation frameworks and rigorous validation of both organoid model predictive accuracy and AI algorithm robustness prior to clinical trial approval[94-96], while ethical concerns center on the integration of sensitive multimodal patient data-spanning genomics, medical imaging, and organoid responses-which raises critical data privacy and security issues[95,97,98], thereby demanding stringent data anonymization, transparent governance protocols, dynamic informed consent processes, and early multidisciplinary dialogue to ensure responsible clinical implementation[95,99,100].

CONCLUSION

The integration of AI and organoid technology represents a paradigm shift in GPL management, moving beyond static diagnosis towards dynamic, mechanism-based precision prevention. To realize this transformative potential, future efforts must prioritize: (1) The establishment of standardized multi-omics and organoid biobanks; (2) The development of explainable and generalizable AI models; and (3) The execution of large-scale, prospective clinical trials to validate this integrative pipeline. Through concerted interdisciplinary collaboration, the synergy of AI and organoids is poised to redefine the future of EGC intervention, ultimately translating deep mechanistic understanding into tangible improvements in patient survival.