Artificial intelligence in functional gastrointestinal disorders: From precision diagnosis to preventive healthcare

Abstract

Functional gastrointestinal disorders (FGIDs), including irritable bowel syndrome (IBS), functional dyspepsia (FD), and gastroesophageal reflux disease (GERD), present persistent diagnostic and therapeutic challenges due to symptom heterogeneity and the absence of reliable biomarkers. Artificial intelligence (AI) enables the integration of multimodal data to enhance FGID management through precision diagnostics and preventive healthcare. This minireview summarizes recent advancements in AI applications for FGIDs, highlighting progress in diagnostic accuracy, subtype classification, personalized interventions, and preventive strategies inspired by the traditional Chinese medicine concept of “treating the undiseased”. Machine learning and deep learning algorithms have demonstrated value in improving IBS diagnosis, refining FD neuro-gastrointestinal subtyping, and screening for GERD-related complications. Moreover, AI supports dietary, psychological, and integrative medicine-based interventions to improve patient adherence and quality of life. Nonetheless, key challenges remain, including data heterogeneity, limited model interpretability, and the need for robust clinical validation. Future directions emphasize interdisciplinary collaboration, the development of multimodal and explainable AI models, and the creation of patient-centered platforms to facilitate a shift from reactive treatment to proactive prevention. This review provides a systematic framework to guide the clinical application and theoretical innovation of AI in FGIDs.

Key Words: Artificial intelligence; Functional gastrointestinal disorders; Irritable bowel syndrome; Functional dyspepsia; Gastroesophageal reflux disease

Core Tip: Artificial intelligence (AI) is reshaping the management of functional gastrointestinal disorders by integrating clinical, physiological, and imaging data for precise diagnosis, refined subtyping, and tailored treatments. Preventive approaches, including those informed by traditional Chinese medicine, show promise in improving patient outcomes. However, data variability and limited model transparency remain key challenges. Advances in interpretable and clinically validated AI will support a shift from reactive treatment to proactive and preventive care.

INTRODUCTION

Functional gastrointestinal disorders (FGIDs), including irritable bowel syndrome (IBS), functional dyspepsia (FD), and gastroesophageal reflux disease (GERD), are highly prevalent chronic conditions characterized by persistent or recurrent gastrointestinal symptoms in the absence of detectable organic pathology. Under the latest Rome IV criteria, FGIDs are classified as disorders of gut–brain interaction, reflecting the complex interplay between the central nervous system, gastrointestinal tract, and psychological factors[1]. FGIDs affect up to 40% of the global population, with a higher prevalence among women, and impose a significant burden on both patients and healthcare systems worldwide[2].

Despite their high prevalence, FGIDs remain challenging to diagnose and manage. Current diagnostic approaches rely primarily on symptom-based criteria, such as Rome IV, often supplemented by laboratory and imaging tests to rule out organic disease[3]. However, these approaches are constrained by reliance on subjective symptom reporting, the absence of reliable biomarkers, and substantial overlap among subtypes, which collectively contribute to diagnostic uncertainty, frequent comorbidities, and delays in implementing targeted management strategies[4,5].

In recent years, artificial intelligence (AI) technologies-particularly machine learning (ML) and deep learning (DL) algorithms-have shown considerable potential in addressing these unmet needs. AI excels at detecting subtle patterns and building predictive models from large, complex datasets that include clinical symptoms, physiological signals, endoscopic images, and multi-omics profiles[6]. Recent studies demonstrate that AI can enhance diagnostic accuracy, refine disease subtyping, and inform treatment decisions for FGIDs by leveraging electronic health records, gastrointestinal motility data, endoscopic imaging, and patient-reported outcomes. Furthermore, the integration of multi-omics data, such as gut microbiota, metabolomics, and neuroimaging-has opened new avenues for identifying novel biomarkers and mechanistic subtypes[7].

Beyond diagnosis, AI-driven predictive modeling supports personalized treatment by anticipating responses to pharmacological and non-pharmacological interventions, including dietary modifications and psychological therapies, thereby advancing precision medicine[8]. With these advances, AI applications in FGIDs are rapidly transitioning from experimental research to early clinical implementation, shaping the future of digital healthcare and digestive disease management.

This review focuses on the applications of AI in the diagnosis, classification, and management of IBS, FD, and GERD. Although GERD presents both functional and structural characteristics, its considerable overlap with other FGIDs-particularly in symptom presentation and gut–brain axis involvement-justifies its inclusion in this context. In this review, AI is defined broadly to encompass ML, DL, and computer vision (CV), with particular emphasis on DL and CV given their central roles in analyzing imaging and multimodal datasets. Although large language models, such as ChatGPT, are increasingly used in medical settings, they remain general-purpose tools for language processing and documentation rather than disease-specific diagnostic aids. Therefore, this review concentrates on AI methodologies directly relevant to clinical analysis and decision-making in FGIDs.

COMMON MECHANISMS AND HOLISTIC PERSPECTIVES OF FGIDS

Although IBS, FD, and GERD exhibit distinct clinical presentations, they share substantial overlap in underlying mechanisms, symptom profiles, and treatment responses. Epidemiological studies report prevalence rates of 11.2% for GERD, 7.7% for FD, and 10.5% for IBS, with approximately 30.7% of affected individuals experiencing comorbidity involving two or more of these disorders[9]. This high rate of overlap is not merely the result of diagnostic imprecision; rather, it suggests deeper mechanistic convergence and possibly evolutionary links. Consequently, an increasing number of researchers advocate a spectrum model to reconceptualize FGIDs, proposing that IBS, FD, and GERD represent clinical phenotypes of a shared pathophysiological network that manifests across different gastrointestinal regions or disease stages[10-12]. This model departs from traditional organ-centric perspectives, emphasizing cross-mechanism interactions, systemic dysregulation, and individual variability in clinical trajectories and treatment responses.

As illustrated in Figure 1, current evidence highlights five interrelated core mechanisms underpinning FGIDs: Gut–brain axis (GBA) dysfunction, gastrointestinal motility and sensory abnormalities, gut microbiota dysbiosis, mucosal barrier impairment with low-grade inflammation, and psychosocial modulation. These mechanisms interact dynamically, forming a complex network that drives the heterogeneity of clinical manifestations.

Figure 1 Integrated systemic mechanisms underlying functional gastrointestinal disorders. HPA: Hypothalamic–pituitary–adrenal; LES: Lower esophageal sphincter; GI: Gastrointestinal.

GBA dysregulation

The GBA mediates bidirectional communication between the central nervous system and the enteric nervous system, playing a pivotal role in sensory regulation, stress responses, and immune signaling[13]. In FGIDs, this axis is frequently dysregulated. Patients commonly exhibit visceral hypersensitivity, characterized by heightened perception of gastrointestinal stimuli in the absence of structural pathology[14]. This heightened sensitivity is associated with aberrant central processing of visceral signals and impaired sensory modulation. Chronic stress often leads to overactivation of the hypothalamic–pituitary–adrenal axis, inducing gastrointestinal motility disturbances and mucosal inflammation, thereby further disrupting gut function[15]. Reduced vagal tone also contributes to impaired top-down regulation of gastrointestinal activity[16]. These neuroendocrine disturbances frequently co-occur with anxiety and depression, forming a well-recognized emotion–gut–symptom amplification pathway in FGIDs.

Gastrointestinal motility and sensory dysfunction

Gastrointestinal motility abnormalities, often accompanied by altered sensory processing, represent core pathological features of FGIDs. Patients with FD frequently present with delayed gastric emptying and impaired gastric accommodation[17], while IBS is often associated with disrupted intestinal peristaltic rhythms and abnormal defecatory dynamics, manifesting as constipation, diarrhea, or alternating patterns[18]. In GERD, lower esophageal sphincter (LES) dysfunction or impaired esophageal clearance facilitates gastric content reflux and mucosal injury. Additionally, FGID patients frequently exhibit lower sensory thresholds, resulting in amplified or painful perceptions even under normal physiological conditions[19]. These intertwined disturbances in motility and sensory regulation create a maladaptive feedback loop that perpetuates symptoms and may contribute to the chronicity of these disorders.

Gut microbial dysbiosis

The gut microbiota plays a critical role in maintaining gastrointestinal homeostasis, including immune regulation, motility, and neural signaling. Recent studies consistently report microbial dysbiosis in FGIDs. IBS patients often exhibit reduced microbial diversity, with specific alterations such as increased methanogenic or gas-producing bacteria linked to diarrhea- or constipation-predominant subtypes[20]. FD has been associated with small intestinal bacterial overgrowth or displacement of antral microbiota[21]. These alterations disrupt gastrointestinal function through mechanisms such as altered short-chain fatty acid production, increased mucosal permeability, and neuroimmune interactions[22]. In GERD, structural changes along the oral–esophageal–intestinal axis suggest that microbial dysbiosis contributes to broader digestive tract dysregulation[23].

Mucosal inflammation and barrier dysfunction

Although FGIDs have traditionally been considered “non-inflammatory”, emerging evidence indicates the presence of low-grade mucosal inflammation and barrier dysfunction. In IBS, for instance, the small intestinal mucosa often exhibits mast cell and T-cell infiltration, accompanied by elevated inflammatory cytokines such as tumor necrosis factor-alpha and interleukin-6[24]. While these changes may not result in overt ulceration or gross structural damage, they are sufficient to activate sensory nerve endings, driving visceral hypersensitivity and pain. Increased epithelial permeability allows luminal antigens and bacterial products to penetrate the lamina propria, amplifying immune activation[25]. Similar processes have been documented in the esophageal mucosa of GERD patients[26]. Collectively, this low-grade inflammation acts as both a mediator of symptom generation and a peripheral driver of GBA dysregulation.

Emotional modulation

Psychological and emotional factors play a pivotal role in the pathogenesis, symptom perception, and therapeutic outcomes of FGIDs. Patients with IBS, FD, and related disorders frequently exhibit higher rates of anxiety, depression, and stress-related conditions compared with the general population[27,28]. These psychological disturbances interact closely with the GBA, influencing visceral sensitivity, gastrointestinal motility, and immune responses, thereby exacerbating symptom severity and clinical heterogeneity[29].

Neuroimaging studies have revealed altered connectivity and activity in emotion-regulating regions, such as the anterior cingulate cortex and insula, reinforcing the conceptualization of FGIDs as prototypical disorders of brain-gut interaction[30]. Additionally, specific cognitive and behavioral traits-such as somatization, catastrophizing, and illness-related stigma-can amplify symptom perception, reduce coping capacity, and influence healthcare-seeking behaviors.

These psychological mechanisms provide a strong foundation for AI-driven models that integrate psychological and physiological data, facilitating more precise phenotyping and tailored interventions aimed at optimizing clinical outcomes.

Holistic perspective and systems thinking

The spectrum model not only provides an integrative framework for understanding FGID mechanisms but also signals a paradigm shift in medical thinking-from reductionist approaches toward systems-based medicine. From this perspective, FGIDs exemplify how localized dysfunction within the gastrointestinal tract can disrupt broader physiological networks: Gastrointestinal dysregulation reverberates across neural, immune, psychological, and metabolic domains, resulting in complex adaptive imbalances[31].

This systems-based understanding aligns with the holistic principles of traditional Chinese medicine (TCM), which conceptualizes the spleen and stomach as the “foundation of acquired constitution” responsible for governing qi and blood production and maintaining systemic homeostasis[32]. Within TCM theory, impaired stomach qi manifests not only as digestive dysfunction but also as systemic deficits, such as reduced immunity and fatigue. Modern research corroborates this “gut–systemic disease linkage” hypothesis, associating gastrointestinal dysfunction with metabolic syndrome, neurodegenerative diseases, and systemic inflammaging[33].

Within this multidimensional dysregulation framework, traditional analytic approaches often fail to capture the intricate interactions among variables. AI provides unprecedented opportunities by integrating multimodal datasets-including electronic health records (EHRs), endoscopic images, physiological signals, microbiomics, and psychological assessments-to uncover hidden patterns, redefine mechanistic subtypes, and predict individualized therapeutic responses through ML models[34–36]. This approach not only advances mechanistic understanding and reclassification of FGIDs but also elevates precision medicine from an organ-centric to a network-based paradigm. Thus, holistic perspectives and systems thinking form both the conceptual foundation for understanding the complexity of FGIDs and the theoretical basis for applying AI to develop mechanism-driven diagnostic and therapeutic strategies.

AI IN FGIDS

AI applications in IBS

IBS, one of the most extensively studied FGID subtypes, is characterized by recurrent abdominal pain and altered bowel habits and is classified into diarrhea-predominant (IBS-D), constipation-predominant, mixed, and unclassified subtypes. However, these subtypes fluctuate over time and symptom-based diagnoses lack objective markers. AI has been increasingly explored for IBS diagnostic support, subtype identification, biomarker discovery, and personalized intervention design.

From symptoms to multimodal data for diagnosis and subtyping: IBS diagnosis primarily relies on the Rome IV symptom criteria and lacks definitive biomarkers. AI offers a novel solution to this problem. A widely cited study utilized abdominal bowel sound signals collected via external microphones and trained ML models to distinguish IBS patients from healthy controls with over 87% accuracy in validation sets[37]. Beyond acoustic signals, AI integrates symptom questionnaires, psychological scores, and EHRs, with some models achieving near-expert level accuracy[38]. Unsupervised clustering approaches have also revealed new, data-driven subgroups that refine traditional IBS subtyping and support personalized treatment strategies[39].

Identifying latent features in mucosal and imaging data: Although IBS shows no organic lesions on routine colonoscopy, DL algorithms have been used to identify microstructural changes in the endoscopic images. For instance, Tabata et al[40] used auto ML to train image recognition models to distinguish IBS from healthy mucosal images with an area under the curve (AUC) of 0.95, indicating that AI can detect biofilm or mucosal microcirculatory differences imperceptible to conventional methods[40]. Emerging physiological signals, such as fecal Volatile Organic Compounds (VOCs), and brain functional imaging [e.g., functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS)] are also viable AI data sources. Studies have shown that VOC profiles and mucosal biofilm quantification of IBS patients can be analyzed by AI models, suggesting that future diagnostics may integrate multiple physiological and microbiome-derived dimensions[41].

Gut microbiota and multi-omics integration: Gut microbial dysbiosis is a key mechanism underlying IBS pathogenesis. AI excels at processing high-dimensional heterogeneous data. Fukui et al[42] developed a classification model based on metagenomic microbial features to achieve robust IBS identification[42]. Similarly, Tanaka et al’s ML model, based on fecal protease profiles, identified specific enzymatic patterns in IBS-D with 90.5% accuracy, surpassing traditional microbial structure models[43]. These findings suggest that AI-driven multi-omics integration (microbiome, metabolomics, and proteomics) can uncover novel IBS subtype biomarkers, advancing mechanistic stratification and precision interventions.

Personalized treatment strategies and industrial applications: IBS treatment hinges on individual responses, with traditional approaches often mired in trial and error, limiting efficacy and adherence. AI introduces precision treatment pathways. A recent multicenter randomized controlled trial compared an AI-assisted personalized diet (PD) with standard low-FODMAP diets and found that both interventions alleviated IBS symptoms within 6 weeks, but the PD group showed superior quality of life improvements and microbial diversity across IBS subtypes[44]. Additionally, the Heali mobile application, a digital tool for low-FODMAP diet guidance, has improved the quality of life and bowel habits of patients with IBS[45].

AI applications in IBS have expanded from diagnostic support to disease subtyping, mechanistic research, and personalized interventions, thereby forming a closed-loop management pathway from identification to treatment. Although most studies remain exploratory, ongoing multicenter clinical validations and advances in cross-modal data integration will help clarify the potential of AI in IBS clinical pathways, with promising prospects for broader clinical implementation.

AI applications in FD

FD is a prevalent FGID characterized by chronic upper abdominal discomfort including postprandial fullness, early satiety, and epigastric pain. Clinically, FD is categorized as postprandial distress syndrome (PDS) and epigastric pain syndrome (EPS). However, over 40% of patients exhibit overlapping PDS and EPS symptoms, challenging the stability and utility of these subtypes[46]. AI applications in FD, spanning symptom-driven reclassification, multimodal physiological signal analysis, and intervention prediction, are reshaping our understanding of FD pathophysiology and treatment paradigms.

Symptom phenotyping and integrative subtyping: Traditional PDS/EPS classification based on subjective symptoms lacks biomarker support and struggles to address clinical heterogeneity. Mousavi et al[47] applied unsupervised clustering to large-scale questionnaires and psychological comorbidity data, identifying FGID patient subgroups that transcend IBS and FD boundaries, presenting mixed phenotypes such as epigastric pain with diarrhea or bloating. This “label-free” approach highlights the spectrum of FGID subtypes and offers new dimensional clues for FD classification. AI has also enhanced TCM pattern differentiation, which refers to the classification of patient presentations into distinct diagnostic “patterns” based on integrated symptomatology and physiological characteristics within TCM theory. Yoon et al[48] trained a random forest model on 21 TCM symptoms in patients with FD and identified three major patterns-spleen-stomach weakness (reflecting impaired digestive and absorptive function), liver-stomach disharmony (associated with stress-related autonomic and gastrointestinal dysregulation), and food stagnation (indicative of delayed gastric emptying and motility dysfunction). Notably, the model prioritized features (e.g., chest oppression and sallow complexion) that differed from expert consensus[48], suggesting that AI can uncover non-intuitive associations that may help standardize TCM diagnostic frameworks. Additionally, a Korean study integrated fNIRS brain activity, skin conductance, and pulse wave data to train AI models aligned with TCM pattern classifications, exploring objective, data-driven approaches to pattern differentiation[49]. These efforts underscore AI’s potential to advance integrative Western–TCM research and to promote more standardized, evidence-based diagnostic strategies.

Brain-gut imaging and gastrointestinal motility analysis: Although FD lacks structural abnormalities, functional changes in the neuro-gastrointestinal regulatory network can be detected using advanced physiological imaging. Katsumata et al[50] used fNIRS signals from the prefrontal cortex during food stimuli, combined with food preference scores, to develop a neural network model for FD diagnosis and subtyping, achieving 77% accuracy in distinguishing FD from healthy controls and supporting the central role of brain-gut interaction in FD[50]. This model also partially differentiates FD from IBS, reinforcing their mechanistic overlap despite clinical distinctions. In endoscopic imaging, Mihara et al’s DL model identified FD in Helicobacter pylori-negative patients from duodenoscopy images, achieving an AUC of 0.85 and generating interpretable heatmaps highlighting mucosal microstructural regions[51]. This study pioneered the identification of FD biological features in “seemingly normal” endoscopic images, positioning AI as a potential endoscopic diagnostic tool.

Gastric motility parameters and treatment response prediction: Gastric motility abnormalities, such as delayed gastric emptying and impaired accommodation, are the hallmark mechanisms of FD, particularly PDS. However, interpretation of gastric emptying scintigraphy (GES) and wireless capsule pressure data relies on subjective expertise. Takakura et al[52] integrated GES data, capsule-derived gastrointestinal transit metrics (e.g., Duodenal Motility Index), and clinical variables [body mass index (BMI), diabetes history, and infection prodrome] into a Ridge Regression model to predict 6-month responses to prokinetics or neuromodulators. In non-diabetic patients with delayed gastric emptying, the model achieved an AUC of 0.83 for predicting prokinetic response, outperforming traditional metrics and enabling prospective “treatment matching”. AI also shows promise in analyzing electrogastrography (EGG) and high-resolution manometry data[53], potentially clarifying boundaries between functional gastroparesis and FD.

AI-assisted digital therapeutics and smart management: Patients with FD frequently experience psychological comorbidities, such as anxiety and sleep disturbances, necessitating non-pharmacological interventions for long-term symptom control. AI-driven digital health tools have emerged as a key adjunct. Smartphone applications enable diet logging, symptom tracking, and cognitive behavioral guidance, with AI models identifying individualized triggers (e.g., specific foods, stress, menstrual cycles) and delivering tailored recommendations[54]. Advanced platforms incorporate augmented reality (AR) and wearable devices to create closed-loop intervention systems. For instance, AR glasses guide diaphragmatic breathing, while AI analyzes heart rate variability and skin conductance to assess relaxation states, providing real-time feedback for personalized training[55]. Such AI-empowered digital therapeutics are becoming a new frontier in FGID management[56].

The clinical heterogeneity of FD challenges traditional “symptom-subtype-treatment” paradigms. AI, through data-driven phenotyping, multimodal imaging analysis, treatment response prediction, and digital management, has progressively redefined the FD disease framework.

Applications in GERD

GERD is characterized by gastric content reflux into the esophagus causing symptoms (e.g., heartburn and regurgitation) or complications [e.g., esophagitis, Barrett’s esophagus (BE), esophageal adenocarcinoma], spanning phenotypes from non-erosive reflux disease to erosive esophagitis and BE. It combines functional components with the risk of organic progression. AI has significant potential in GERD management, encompassing automated diagnostics, mucosal lesion detection, treatment decision prediction, and personalized patient stratification.

Intelligent interpretation of pH-impedance monitoring: 24-hour pH-impedance monitoring is a cornerstone for GERD diagnosis, yet its interpretation is labor-intensive and expertise-dependent. Zhou et al[57] developed a DL system for automated reflux event detection, achieving an AUC of 0.87, surpassing commercial software (AUC 0.40) and approaching expert performance (AUC 0.83). This system identifies acidic and non-acidic reflux and calculates key metrics, such as acid exposure time and symptom association probability, to enhance report quality and consistency. Emerging parameters, such as low baseline impedance and post-swallow peristaltic wave index (PSPW), predict proton pump inhibitor (PPI) efficacy and assess esophageal mucosal function[58]. The manual calculation of these metrics is time consuming and subjective. Wong et al[59] reported that AI models accurately identified PSPW events with an intraclass correlation coefficient of 0.921 and 82% sensitivity compared to manual evaluation, offering a pathway for standardized pH reporting[59].

Endoscopic image recognition and early BE screening: AI applications in endoscopic image recognition have advanced GERD management, particularly for BE and its precancerous lesions. Fockens et al[60] developed a computer-aided detection (CADe) system achieved 90% sensitivity for early Barrett-related neoplasia, outperforming general endoscopists (74%) and matching expert levels. In simulated clinical settings, CADe increased endoscopists’ detection rates from 67% to 79% without compromising specificity, demonstrating AI’s role as a “second observer” for high-risk lesions[61]. AI also aids in assessing LES anatomy, such as gastroesophageal flap valve Hill grading. A ResNet-50 model trained on over 3000 images achieved 93.4% accuracy and an AUC of 0.989 for Hill classification, surpassing those of senior endoscopists[62]. This capability supports the evaluation of LES integrity and GERD severity, identifying anatomical anomalies (e.g., hiatal hernia) to guide treatment.

AI differentiation of functional heartburn and true GERD: Approximately 30%-40% of heartburn patients lack abnormal acid exposure on objective testing, indicating functional heartburn, which often responds poorly to PPIs and risks misdiagnosis. AI integrates multidimensional features-symptom patterns, pH-impedance results, and psychological states-to distinguish functional heartburn from true GERD[63]. Low PSPW and baseline impedance indicate mucosal barrier disruption and impaired clearance, predicting favorable PPI responses, whereas patients with functional heartburn typically show normal values[64]. Some AI models have developed GERD subtyping algorithms, classifying patients into acid-driven, pain-sensitized, or motility-disordered subtypes, enabling tailored interventions, such as PPIs, pain modulators, prokinetics, or psychotherapy[65]. This offers a mechanism-driven approach to refractory GERD management.

GERD surgery and treatment response prediction: Anti-reflux surgery (e.g., Nissen fundoplication) is a common option for PPI-refractory GERD, but postoperative outcomes vary widely. AI models that integrate clinical data (e.g., esophageal manometry, pH monitoring, and symptom scales) predict surgical efficacy. Incorporating variables such as BMI, LES pressure, and IBS comorbidity, these models identify high-risk groups with a lower likelihood of postoperative relief, aiding surgeons in personalized preoperative assessments[66]. In BE management, AI shows promise for early prediction of neoplasia. By integrating histopathological data, molecular biomarkers (e.g., DNA methylation status), and endoscopic features, AI models can forecast cancer risk and guide follow-up and intervention strategies[67].

AI has substantial potential for GERD diagnosis, subtyping, and management. Coupled with wearable devices and remote monitoring, AI enables the real-time tracking of nocturnal or positional reflux. Self-management applications that leverage daily symptoms, diet, and behavioral data can build personalized reflux prevention models. Through these advancements, GERD management has become more proactive, continuous, and precise.

Commonalities and differences in cross-disease AI models

AI demonstrates substantial potential in managing FGIDs, particularly IBS, FD, and GERD, by leveraging multimodal data integration and precise modeling to enhance diagnosis, subtyping, and treatment prediction. However, the distinct pathophysiological mechanisms, clinical needs, and data characteristics of these disorders shape both the shared features and specific requirements in AI model design and application. Table 1 provides a comparative overview of AI applications in IBS, FD, and GERD, highlighting the key similarities, unique challenges, and model considerations. This section builds on this comparison to analyze commonalities, disease-specific demands, and the potential and challenges of cross-disease models, offering directions for future research.

Table 1 Comparison of artificial intelligence applications in irritable bowel syndrome, functional dyspepsia, and gastroesophageal reflux disease.

Aspect	IBS	FD	GERD
Common algorithm types	Random Forests and SVM for symptom/microbiome classification; CNN and AutoML for bowel sound and imaging analysis; unsupervised clustering for subtype delineation	Random Forests for TCM pattern identification; unsupervised clustering for symptom subtyping; neural networks for brain-gut imaging; CNN for endoscopic imaging; Ridge Regression for motility prediction	DL (e.g., CNN, ResNet-50) for endoscopic imaging and pH-impedance analysis; ML classification for symptom subtyping and treatment prediction
Primary input data	Symptom questionnaires (Rome IV), psychological scales, EHRs, bowel sound audio, microbiome data, endoscopic images, fNIRS, VOCs	Symptom questionnaires (epigastric pain, postprandial fullness), psychological scales, fNIRS, gastric motility data (GES, manometry), endoscopic images (duodenal), skin conductance, pulse wave	24-hour pH-impedance monitoring, symptom questionnaires, endoscopic images, esophageal manometry, histopathology, DNA methylation biomarkers
Model Objectives	Differentiate IBS from controls (87% accuracy); identify subtypes and novel clusters; predict dietary/drug responses	Differentiate FD from controls (77% accuracy); refine PDS/EPS or TCM patterns; predict prokinetic response (AUC 0.83); identify triggers	Differentiate GERD from functional heartburn (AUC 0.87); detect BE (90% sensitivity); predict PPI/surgical efficacy; assess LES (93.4% accuracy)
Model generalizability	Symptom-psychological clustering applicable to other FGIDs; microbiome and bowel sound models are more specific	Symptom-psychological clustering and brain-gut analysis applicable to IBS; gastric motility models relevant to gastroparesis; TCM models more specific	BE detection models highly specific; pH-impedance and symptom subtyping applicable to other reflux disorders; LES models relevant to motility disorders
Clinical translation progress	Bowel sound diagnosis (87% accuracy)[37]; microbiome models validated[42,43]; digital therapeutics (e.g., Heali App) in trials[44,45]	Early-stage: FNIRS diagnosis (77% accuracy)[50]; endoscopic imaging models (AUC 0.85)[51]; digital therapeutics (e.g., diet apps, AR breathing) in validation[54,55]	More mature: CADe endoscopic systems (90% sensitivity) in trials[60]; pH-impedance analysis (AUC 0.87) near commercialization[57,59]; surgical prediction models validated[66]

AI: Artificial intelligence; IBS: Irritable bowel syndrome; FD: Functional dyspepsia; GERD: Gastroesophageal reflux disease; SVM: Support vector machine; CNN: Convolutional neural network; AutoML: Automated machine learning; TCM: Traditional Chinese medicine; DL: Deep learning; ML: Machine learning; EHRs: Electronic health records; fNIRS: Functional near-infrared spectroscopy; VOCs: Volatile organic compounds; GES: Gastric emptying scintigraphy; GERQ: Gastroesophageal reflux questionnaire; NBI: Narrow-band imaging; PDS: Postprandial distress syndrome; EPS: Epigastric pain syndrome; AUC: Area under the curve; BE: Barrett’s esophagus; PPI: Proton pump inhibitor; LES: Lower esophageal sphincter; FGIDs: Functional gastrointestinal disorders; AR: Augmented reality; CADe: Computer-aided detection.

AI applications for IBS, FD, and GERD share several commonalities. First, multimodal data integration is central and encompasses symptom questionnaires, psychological scores, and physiological signals. For instance, clustering analyses of symptoms and psychological data have redefined subtypes of IBS and FD, suggesting a disease spectrum[39,47]. Second, algorithmic diversity is evident, with ML (e.g., Random Forests, Support Vector Machines) and DL (e.g., Convolutional Neural Networks) being widely applied for symptom classification, image recognition, and motility modeling. DL excels in IBS endoscopic image analysis (AUC 0.95)[40] and GERD LES assessment (93.4% accuracy)[62]. Third, the clinical goals focus on precise diagnosis, subtype refinement, and treatment prediction, reflecting a shared trend toward personalized medicine. Examples include IBS dietary intervention prediction[44], FD prokinetic response prediction (AUC 0.83)[52], and GERD PPI efficacy assessment[64]. The brain-gut axis perspective further unifies applications, with fNIRS and psychological data revealing neural dysregulation in IBS and FD[50], and aiding functional heartburn subtyping in GERD[63]. These commonalities suggest that cross-disease AI models can achieve generalizable predictions (potentially AUC > 0.80) using shared data architectures and algorithms.

However, disease-specific characteristics drive notable differences in the model design. Data sources vary: IBS relies on non-invasive signals (e.g., bowel sounds, microbiome) to address diagnostic exclusion of organic disease[37,42], FD emphasizes brain-gut imaging and gastric motility data to capture functional dysregulation[50,52], and GERD prioritizes pH impedance and endoscopic imaging to address both functional and organic risks[57,60]. The model objectives differ: IBS focuses on symptom heterogeneity and dietary interventions[39,44], FD on subtyping and TCM pattern exploration[47,48], and GERD on BE detection and functional heartburn differentiation[60,63]. Clinical translation maturity also varies: GERD’s endoscopic detection systems are in multicenter trials (90% sensitivity)[60], IBS digital therapeutics are commercialized[45], and FD remains in early validation[50,51]. Regarding generalizability, IBS and FD symptom-psychological clustering shows cross-FGID potential[39,47], but GERD BE detection is highly specific, limiting direct transferability[60]. These differences highlight the need for cross-disease models that balance universal features with disease-specific demands.

Developing cross-disease AI models has significant potential but faces challenges. Symptom-psychological data and microbiota are key integration points, with the former predicting FGID subtypes[47] and the latter validated in IBS, with exploratory promise in FD and GERD[42,67]. However, data heterogeneity demands standardized protocols and advanced multimodal fusion techniques, which are currently limited in maturity[68]. The tension between model specificity (e.g., GERD endoscopic analysis) and generalizability requires modular design. Clinical validation progresses faster in IBS and GERD[44,60] but lags in FD due to insufficient multicenter data[50]. Data privacy and ethical concerns necessitate solutions, such as federated learning, to comply with regulations. Future cross-disease models could adopt multimodal frameworks integrating symptoms, imaging, and omics data, leveraging transformer-based architectures for enhanced generalizability while developing interactive tools to improve clinical interpretability[69].

AI applications in FGIDs reveal deep connections between gastrointestinal and systemic health, particularly through microbiota and brain-gut axis data, suggesting that gut function influences immunity, metabolism, and aging processes[42,50]. These insights lay the groundwork for exploring AI’s broader role in maintaining health. The next section focuses on the intersection of gastrointestinal health and anti-aging, integrating Western and TCM perspectives to examine how AI extends beyond disease management to prevention and the enhancement of systemic vitality.

AI-ENABLED GASTROINTESTINAL HEALTH AND PREVENTIVE MEDICINE

Gastrointestinal health is pivotal not only to digestive function but also to systemic homeostasis of immunity, metabolism, and the nervous system. TCM posits “stomach qi” as the foundation of vitality, a concept that aligns with modern research on the regulatory role of the gut microbiome in health. AI, by integrating multi-omics data (genomics, proteomics, metabolomics), clinical records, and physiological signals, offers unprecedented tools for personalized health management. This section explores how AI reshapes FGID management by assessing gut health, identifying disease risks, and promoting preventive strategies, transitioning from disease treatment to the TCM principle of “treating the undiseased” (preventive care).

Integrative western-TCM perspectives: Gastrointestinal health as a systemic vitality nexus

TCM emphasizes “stomach qi as the foundation”, viewing gastrointestinal function as the cornerstone of acquired constitution, directly influencing overall health and disease prevention. Modern research has revealed that the gut microbiome in healthy individuals exhibits greater diversity and anti-inflammatory properties and is enriched with probiotics (e.g., Bifidobacterium), while dysbiosis is linked to immune decline and chronic inflammation[70]. These findings resonate with TCM’s holistic perspective of TCM, which positions gut health as a critical hub for systemic vitality. AI bridges TCM’s empirical wisdom of TCM with modern systems biology, offering a cross-disciplinary lens for FGID management. For instance, AI analyzes interactions between the gut microbiome and the neuro-immune axis, elucidating the underlying mechanisms in IBS, FD, and GERD[42] and providing a scientific basis for preventive care.

AI-driven gut health assessment and risk identification

AI’s capacity to process high-dimensional, heterogeneous data makes it ideal for assessing gut health and identifying disease risks. Studies have demonstrated that AI models can construct quantitative metrics from fecal microbiome data to dynamically evaluate individual health status, with an average error of approximately four years[71]. For example, a DL model predicted host age from microbial taxonomic profiles, achieving a mean absolute error of 5.91 years on external datasets[72]. Additionally, the gut microbiome health index (GMHI), derived via ML, integrates beneficial and pathogenic bacterial ratios to generate a score distinguishing healthy individuals from those with chronic diseases[73]. In FGIDs, AI detects dysbiosis patterns (e.g., microbial imbalances in IBS)[39] and predicts symptom recurrence or complication risks, thus providing quantifiable biomarkers for early intervention. These models shift diagnostics from reactive to proactive prevention, embodying the essence of “treating the undiseased”.

AI-empowered preventive health management

AI supports FGID management within preventive health paradigms by designing personalized interventions through multimodal data integration. AI-optimized dietary interventions have been shown to enhance microbiome diversity, with studies reporting a 20% increase after intervention[74]. Wearable devices (such as gut acoustic sensors) and smart home technologies (such as intelligent toilets) have expanded AI’s monitoring capabilities and can deliver real-time gut health reports via image recognition and biochemical analysis[75]. In FGIDs, AI can predict symptom relapse risk, for example by monitoring microbial deviations and stress indices in IBS patients, to recommend preemptive dietary or psychological interventions[44]. These strategies, enabled by continuous monitoring and early intervention, embody the concept of “treating the undiseased” and contribute to an extended healthspan.

The overall framework of AI-enabled precision management and preventive healthcare for FGIDs is illustrated in Figure 2. In this conceptual model, multimodal health data-including clinical symptoms, stool microbiome, wearable sensor outputs, physiological indicators, and dietary habits-are integrated by advanced AI models to support dynamic health assessment, risk prediction, and proactive, personalized interventions. Such interventions encompass dietary, pharmacologic, psychological, and TCM-based therapies, which are delivered and monitored via digital platforms. This closed-loop, cross-disciplinary management strategy exemplifies the shift from passive treatment to proactive prevention in FGID care.

Figure 2 Artificial intelligence driven workflow for precision management and prevention of functional gastrointestinal disorders, integrating multimodal data, digital platforms, and personalized interventions including Western and traditional Chinese medicine. TCM: Traditional Chinese medicine.

Cross-disciplinary integration: AI bridging traditional and modern medicine

AI has excelled at integrating both Western and TCM data and perspectives. For example, studies have correlated fecal microbiome metrics with TCM tongue coating and pulse patterns, identified dysbiosis patterns corresponding to abnormal tongue signs, and provided data-driven evidence for integrative medicine[76]. Knowledge graph technologies further enhance AI’s integrative capacity by constructing networks that encompass dietary nutrition, microbial metabolism, immune responses, and TCM dietary therapy principles, supporting the development of interventions that blend TCM food therapy with Western nutritional science[77]. In FGIDs, this cross-disciplinary approach helps optimize IBS dietary management, FD pattern stratification, and GERD complication prevention[50,60]. By merging TCM’s holistic perspective with modern systems biology, AI enhances the scientific rigor and cultural adaptability of health management, offering a new paradigm for global health practices.

LIMITATIONS AND FUTURE DIRECTIONS OF AI IN FGIDS MANAGEMENT

AI holds transformative potential in FGID management; however, its clinical translation faces significant challenges, including data limitations, insufficient interpretability, and barriers to clinical integration. By systematically analyzing these limitations and proposing refined future directions, this section aims to provide a clear roadmap for advancing AI in FGIDs and maintaining academic rigor and logical coherence.

Current limitations

Despite rapid advances in AI applications for FGIDs, significant challenges remain that limit their translation into routine clinical practice. Current evidence is often constrained by small sample sizes, heterogeneous study designs, and variable outcome definitions, leading to inconsistent results and limited generalizability across populations. Many studies report promising performance metrics, but these are frequently derived from retrospective, single-center datasets and lack external validation, raising concerns about reproducibility and overfitting. Furthermore, the black-box nature of many DL models undermines trust, while regulatory, ethical, and integration barriers continue to slow clinical adoption. Below, we critically examine these challenges in detail, highlighting areas that require methodological rigor and collaborative efforts to ensure AI tools for FGIDs are both reliable and clinically meaningful.

Data heterogeneity and limited generalizability: The diversity of FGID-related data (such as symptom questionnaires, microbiome profiles, and imaging) enriches AI modeling but also presents substantial challenges. Many studies are limited by single-center, small-sample datasets; for example, IBS models often rely on cohorts of fewer than 500 cases with limited cross-population validation, resulting in poor generalizability[39]. Differences in data standards and sequencing platforms across institutions further complicate model transferability[78]. This heterogeneity undermines model robustness and hampers multicenter collaborative research, highlighting the need for standardized protocols and large-scale datasets to enhance the applicability of AI.

Model interpretability and trust bottlenecks: AI models, particularly DL, are often criticized for their “black-box” nature, making it difficult for clinicians and patients to elucidate decision-making processes, thus eroding trust and ethical compliance[79]. For example, FD endoscopic image analysis models require tools such as heatmaps to highlight the relevant mucosal features to gain clinical acceptance[51]. A lack of interpretability may obscure model biases or errors, thereby compromising diagnostic reliability. Clinicians and patients demand high transparency in FGIDs, where subjective symptoms dominate. While interpretable AI offers effective noninvasive decision support for early gastrointestinal diagnosis, current techniques cover only approximately 30% of cases, requiring further development[80].

Clinical validation and integration challenges: The clinical translation of AI models has been hindered by insufficient validation and integration barriers. Many models, such as those for GERD pH-impedance analysis (AUC 0.87), perform well in controlled settings but may not be as effective in complex clinical environments due to factors such as patient adherence[57]. Additionally, AI as a medical device requires rigorous regulatory approval from agencies such as the Food and Drug Administration (FDA) in the United States or the European Medicines Agency in Europe, involving assessments of performance, safety, and ongoing monitoring-processes that are still evolving[81]. Notably, regulatory frameworks vary across regions: For example, the United States emphasizes real-world evidence and post-market surveillance, the European Union enforces stringent patient privacy and risk classification under General Data Protection Regulation (GDPR) and the AI act, and China has rapidly established AI healthcare regulations focusing on clinical safety and standardized validation. Clinically, physicians need training to adopt AI-assisted decision-making, yet most lack practical experience and relevant knowledge, despite recognizing AI’s growing role[82]. Patient skepticism toward “machine diagnosis” also requires effective communication. These factors collectively delay the integration of AI into FGID clinical practice.

Future directions

Multimodal and cross-disciplinary integration: To address data heterogeneity, future efforts should focus on developing multimodal AI models that integrate symptom, imaging, omics, and wearable device data to enhance diagnostic and therapeutic prediction accuracy. For instance, models combining pre- and post-treatment clinical features with fMRI data can predict acupuncture response in FD[83]. From a cross-disciplinary perspective, AI can standardize TCM pattern differentiation (e.g. “spleen-stomach weakness”) and integrate it with Western biomarkers, developing interventions that are both culturally adaptive and scientifically robust[48]. Techniques such as federated learning enable models to leverage multicenter data while preserving privacy and improving performance and generalizability[84].

Enhancing interpretability and clinical collaboration: Improving model interpretability is critical for AI adoption. Future developments should prioritize explainable AI techniques, such as feature importance analysis and decision pathway visualization, to ensure that clinicians can understand and validate AI recommendations[85]. Interactive dashboards using tools like shapley additive explanations (SHAP) values can elucidate diagnostic rationales, as demonstrated in modeling nonlinear interactions for prostate cancer prognosis[86]. In addition, AI should foster a collaborative ecosystem with clinicians, generating preliminary recommendations for physician refinement. Studies show that such human-AI collaboration can improve diagnostic accuracy, for example, by 15% in acute respiratory distress syndrome[87]. This synergy balances technological efficiency with clinical judgment.

Patient empowerment and ethical assurance: The future of AI goes beyond technical optimization to include patient empowerment and ethical compliance. Patient-facing health management platforms, such as symptom-tracking applications, can improve adherence through personalized recommendations; studies have reported adherence improvements of 7%-40%, supporting home-based medication management[88]. In FGIDs, these tools allow patients to proactively adjust their diet or psychological state, thereby reducing the burden on healthcare systems. Ethically, compliance with regulations such as the GDPR, along with data anonymization and fairness assessments, helps mitigate privacy risks and model biases[89]. Following FDA guidelines for AI medical devices and establishing continuous monitoring mechanisms to adapt to changes in patient populations and disease spectra will help accelerate the compliant deployment of AI tools[90].

Despite notable progress, several interconnected challenges continue to hinder the clinical translation of AI in FGIDs. Small sample sizes and limited cross-cohort validation restrict model generalizability and external applicability. Data heterogeneity across centers-from inconsistent clinical annotations to variations in sequencing platforms and imaging protocols-exacerbates these limitations. Model opacity, especially in DL systems, further undermines clinical trust and regulatory approval, while ethical and privacy concerns, including data security and bias mitigation, remain significant barriers. To address these gaps, future research should prioritize the development of large, multicenter, and harmonized datasets; implement prospective and longitudinal validation studies; advance explainable and transparent model architectures; and establish clear, region-specific regulatory and ethical frameworks. A collaborative, interdisciplinary approach that integrates clinicians, data scientists, and policymakers will be essential for building reliable, equitable, and clinically actionable AI tools for FGIDs.

Clinical practice pathways: From precision medicine to preventive healthcare

AI applications in FGID management require systematic pathways to transition from precision diagnostics to preventive health care. Drawing on AI applications in IBS, FD, and GERD, this section proposes a three-stage practice pathway: Technical optimization, clinical validation with interaction refinement, and integration with preventive management. Tailored to FGIDs’ reliance on subjective symptoms and multimodal data, this pathway integrates Western and TCM perspectives to advance AI-enabled “treating the undiseased” practices, offering an actionable framework for clinicians, researchers, and patients.

Technical optimization

Building precise and transparent AI tools: Technical optimization is foundational for the clinical translation of AI in FGID management. This requires multimodal models that integrate symptom questionnaires, physiological signals (such as bowel sounds and pH-impedance), microbiome profiles, and brain imaging data to enhance diagnostic and intervention precision. In IBS, ML algorithms such as support vector machines and random forest have achieved 100% specificity and 98.5% accuracy, respectively[91]. AI analysis of the correlation between bowel sounds and IBS symptoms has enabled the development of non-invasive diagnostic tools, providing clinicians with convenient support[92]. AI also shows promise in microbiome-based dietary design for personalized treatment. A randomized controlled trial found that AI-assisted tailored diets for functional constipation outperformed traditional therapies, with 88% of patients reporting improvements in quality-of-life scores exceeding 50%[93]. In IBS, AI-assisted diets reduced IBS severity scoring system (IBS-SSS) scores by 124.6 ± 28.5 points compared to 31.3 ± 4.2 for standard diets[44]. Similar multimodal approaches can improve diagnostics and interventions for FD and GERD by analyzing symptom patterns or pH-impedance data.

Standardized data management is critical to address data heterogeneity. Unified formats and sharing protocols can integrate heterogeneous data sources and improve the reliability of clinical decisions[6]. Interpretability is also essential, with algorithms such as local interpretable model-agnostic explanations and SHAP providing feature importance insights and enhancing trust in predictions[94]. These efforts aim to develop technically mature, precise, and transparent AI tools to lay the groundwork for clinical validation.

Clinical validation and interaction optimization

Ensuring performance and user experience: AI models require multicenter clinical trials to validate their performance across diverse populations and FGID subtypes, ensuring reliability in real-world clinical settings. In IBS, AI-assisted PD plans have been validated in multiple studies. A multicenter randomized controlled trial (121 patients; 70 in the PD group, 51 in the low-FODMAP group) showed significant IBS-SSS score reductions in the AI-assisted group, with 75% of patients willing to adhere long-term compared with 12.5% in the low-FODMAP group[44]. This study also noted reduced anxiety and improved quality of life across IBS subtypes. Interaction optimization enhances the user experience. Clinicians can access diagnostic rationales via visualization dashboards, and studies have shown increased acceptance[95]. Patient-facing mobile applications offering symptom tracking and personalized recommendations (e.g., dietary adjustments and diaphragmatic breathing) improve adherence, with trials reporting up to 30% increases[96]. Validation must generate peer-reviewed evidence to ensure model reliability in FGID management and support regulatory approvals (e.g., FDA, National Medical Products Administration)[97]. This stage, projected to span two to three years, aims to establish the clinical credibility of AI models.

Integration and preventive management

Achieving holistic healthcare: The final stage integrates validated AI tools into clinical practice and patients’ daily lives to build a prevention-oriented management ecosystem. In hospitals, AI can seamlessly enhance workflows, such as through AI-assisted endoscopic systems that improve diagnostic efficiency and rule out organic diseases in IBS patients[98]. Smart applications, such as the Heali App, paired with wearable devices (e.g., gut acoustic sensors), provide real-time symptom feedback and recommend dietary or psychological interventions to alleviate symptom burdens[45]. At the community level, AI integrates symptoms, microbiome, and lifestyle data to predict disease recurrence risks, embodying “treating the undiseased”. For instance, AI can analyze the microbial deviations and stress indices of patients with IBS to recommend preventive interventions, increase beneficial bacteria, and reduce relapse rates[99]. Ethical compliance with data privacy regulations (e.g., GDPR) through anonymization and fairness assessments prevents data breaches and biases. Continuous monitoring mechanisms dynamically update models to adapt to changes in the population and the disease spectrum. This stage extends FGID management from clinical settings to holistic, lifelong care encompassing at-risk and affected populations.

CONCLUSION

AI has ushered in a new era in the management of FGIDs, driving advances in precision diagnostics, personalized interventions, and patient empowerment that collectively improve quality of life and clinical outcomes. By integrating multimodal data, AI enables the identification of subtle biological patterns in IBS, FD, and GERD-such as bowel sound signatures, mucosal microstructural changes, and microbial dysbiosis-thus bringing greater objectivity and quantifiability to diagnostic processes that have traditionally relied on subjective symptom assessment. In parallel, digital therapeutics and smart monitoring platforms enhance patient adherence and support comprehensive management strategies, ranging from dietary modification to psychological interventions. By incorporating both Western and TCM perspectives, AI supports not only disease treatment but also systemic health management, reflecting the TCM principle of “treating the undiseased”, and reshaping FGID care paradigms toward integrated, holistic models.

Despite these promising developments, the translation of AI into real-world clinical practice requires ongoing validation and refinement. Persistent challenges-including data heterogeneity, limited interpretability, and ethical considerations-necessitate a balance between technological innovation and humanistic care to ensure that AI models are safe, transparent, and patient-centered. Looking forward, enhanced AI–clinician collaboration will foster an intelligent, responsive, and compassionate healthcare ecosystem. Wearable devices and patient engagement platforms will enable real-time health monitoring and personalized interventions, while advanced multimodal models will improve predictions of disease trajectories and treatment responses. Moreover, cross-disciplinary integration will bridge precision medicine with holistic health paradigms, and AI solutions that combine Western approaches with culturally adaptable TCM principles may facilitate broader community-level health promotion.

Ultimately, AI’s role in FGID management will expand beyond traditional clinical settings to become a cornerstone of a holistic, lifelong healthcare framework. By connecting hospitals, communities, and individuals, AI will not only support precise diagnostics but also enable proactive, preventive strategies, safeguarding gastrointestinal health and promoting systemic well-being.