BPG is committed to discovery and dissemination of knowledge
Systematic Reviews Open Access
Copyright: ©Author(s) 2026. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) license. No commercial re-use. See permissions. Published by Baishideng Publishing Group Inc.
Artif Intell Gastroenterol. Aug 8, 2026; 7(2): 123751
Published online Aug 8, 2026. doi: 10.35712/aig.123751
Artificial intelligence and machine learning applications in abdominal tuberculosis diagnosis: A scoping review and translational roadmap
Kumar Kaushik, Department of General Surgery, Rohilkhand Medical College and Hospital, Bareilly 243006, Uttar Pradesh, India
Jyoti Pathania, Department of Anesthesia, Rohilkhand Medical College and Hospital, Bareilly 243006, Uttar Pradesh, India
Pritish Kumar Singh, Department of Gastroenterology, Rajiv Gandhi Cancer Institute and Research Centre, New Delhi 110085, Delhi, India
Monika Dinkar, Department of Obstretrics and Gynaecology, Bhagwan Mahavir Hospital, New Delhi 110034, Delhi, India
Vanya Pathania, Department of Gynaecology and Obstretics, Maharishi Markandeshwar Medical College and Hospital, Solan 173229, Himachal Pradesh, India
ORCID number: Kumar Kaushik (0009-0007-3693-3362); Jyoti Pathania (0000-0002-6674-5174); Pritish Kumar Singh (0009-0008-1029-5582); Monika Dinkar (0009-0005-5726-1898).
Co-first authors: Kumar Kaushik and Jyoti Pathania.
Author contributions: Kaushik K performed the literature search, screened the records, extracted data, conducted the risk of bias assessment, and drafted the initial manuscript; Kaushik K and Pathania J contributed to the study conception and design, and they contributed equally to this manuscript and are co-first authors; Pathania J provided overall supervision, guided the methodology, and critically revised the manuscript for important intellectual content; Singh PK contributed to the clinical interpretation of data related to gastrointestinal tuberculosis and provided critical revisions; Dinkar M assisted with data extraction and performed the statistical analysis; Pathania V assisted in data interpretation and verification. All authors read and approved the final version of the manuscript.
AI contribution statement: AI tool usage is disclosed as follows: Grok by xAI (accessed April 2026) was used for preliminary literature record identification and screening prioritization during the search phase. The initial draft schematic diagrams were generated by Google AI Gemini. Subsequently, the figures for the article were redrawn, so the current images are all hand-drawn by the authors. Grammarly (version accessed April 2026) was employed for language editing, grammar correction, spelling, and punctuation. AI involvement in manuscript preparation was strictly limited to these assistive functions. All screening decisions, data extraction, risk of bias assessment, thematic synthesis, critical interpretation, manuscript drafting, and revision were performed independently by the authors. No AI-generated text was incorporated verbatim into the final manuscript.
Conflict-of-interest statement: All the authors report no relevant conflicts of interest for this article.
PRISMA 2009 Checklist statement: The authors have read the PRISMA 2009 Checklist, and the manuscript was prepared and revised according to the PRISMA 2009 Checklist.
Corresponding author: Jyoti Pathania, MD, Full Professor, Head, Department of Anesthesia, Rohilkhand Medical College and Hospital, Pilbhit Bypass Road, Bareilly 243006, Uttar Pradesh, India. pathaniajyoti7@gmail.com
Received: May 28, 2026
Revised: June 12, 2026
Accepted: June 30, 2026
Published online: August 8, 2026
Processing time: 71 Days and 1.6 Hours

Abstract
BACKGROUND

Abdominal tuberculosis (ATB) remains a major diagnostic challenge in high-burden low-income and middle-income countries due to nonspecific symptoms, paucibacillary disease, low microbiological confirmation rates, and mimicry of Crohn’s disease and other conditions. Artificial intelligence (AI) and machine learning have shown promise in pulmonary tuberculosis but their role in ATB diagnosis is still emerging.

AIM

To map the existing evidence on AI applications for ATB diagnosis, and propose a translational roadmap for equitable clinical integration.

METHODS

This Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews-compliant scoping review searched PubMed, Cochrane Library, and Google Scholar up to April 2026. Two independent reviewers screened records and extracted data. Risk of bias was assessed using adapted Prediction Model Risk of Bias Assessment Tool principles. Studies were thematically categorized and narratively synthesized.

RESULTS

Seventeen studies (16 original retrospective studies and one systematic review) were included. Fourteen studies focused on differentiating intestinal tuberculosis from Crohn’s disease using radiomics, endoscopic deep learning, histopathology, and multimodal approaches, with area under the curves ranging from 0.80-0.96. Multimodal and explainable models performed best. Major gaps included the absence of management-focused applications, predominance of composite reference standards, limited geographic diversity (87.5% from China), small ATB sample sizes, and minimal attention to human immunodeficiency virus status or drug-resistant tuberculosis.

CONCLUSION

AI and machine learning demonstrate strong potential to improve abdominal tuberculosis differentiation from mimics; however, prospective validation in diverse low-income and middle-income countries populations and integration with existing diagnostic frameworks remain urgently needed for equitable impact.

Key Words: Abdominal tuberculosis; Artificial intelligence; Machine learning; Radiomics; Deep learning; Scoping review; Translational roadmap; Crohn disease

Core Tip: This is the first dedicated Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews-compliant scoping review of artificial intelligence and machine learning for abdominal tuberculosis diagnosis and differential diagnosis. It demonstrates high accuracy of radiomics, endoscopic convolutional neural networks, and multimodal models in differentiating abdominal tuberculosis from Crohn’s disease and other mimics, yet reveals a complete absence of management-focused artificial intelligence applications, and predominance of composite reference standards. The proposed four-phase roadmap provides a practical, equity-centered pathway from retrospective evidence to real-world deployment, prioritizing microbiological reference standards, bias audits, infrastructure-appropriate tools, and explicit integration with existing hierarchical diagnostic criteria to reduce diagnostic delays and unnecessary interventions in high-burden settings.



INTRODUCTION

Abdominal tuberculosis (ATB) encompasses intestinal, peritoneal, nodal, and solid-visceral forms and remains a major diagnostic and therapeutic challenge, particularly in high-burden low-income and middle-income countries (LMIC)[1]. According to the World Health Organization Global Tuberculosis Report 2025[2], an estimated 10.7 million people developed tuberculosis in 2024, with 1.23 million deaths; the South-East Asia, African, and Western Pacific regions account for the large majority of cases, and eight countries (led by India at approximately 25% of the global total) contribute approximately two-thirds of the burden. India’s incidence rate has declined to 187 per 100000 population, yet it continues to carry the world’s highest absolute caseload[3]. Extrapulmonary tuberculosis constitutes 15%-20% of notified tuberculosis cases globally, while abdominal tuberculosis specifically accounts for 1%-3% of all tuberculosis cases and 6%-13% of extrapulmonary tuberculosis, with a higher relative burden in endemic LMIC settings such as the Indian subcontinent[4]. Nonspecific clinical presentations, paucibacillary disease, low microbiological confirmation rates, and radiological or endoscopic mimicry of Crohn’s disease, peritoneal carcinomatosis, lymphoma, and other conditions frequently lead to diagnostic delays, inter-observer variability, and inappropriate therapy, including unnecessary surgery in approximately 15% of cases[5].

Evidence-based hierarchical criteria, such as those proposed by Jha et al[5] and operationalized in India’s - Indian Council of Medical Research (ICMR) standard treatment workflow (STW) for the management of adult ATB (2022)[6], integrate clinical, radiological, endoscopic, histopathological, and microbiological findings [including ascitic adenosine deaminase > 39 U/L, caseating granulomas, and mandatory human immunodeficiency virus (HIV) testing]. Yet resource constraints in endemic settings often limit their application, creating opportunities for artificial intelligence (AI) to support objective, rapid decision-making at key STW nodes (e.g., when diagnosis is unclear or response assessment at 4/8 weeks). AI and machine learning (ML) technologies have already proven transformative in pulmonary tuberculosis through computer-aided detection systems that improve sensitivity, objectivity, and workflow efficiency[7]. These same technologies offer objective, rapid, multimodal solutions capable of reducing subjectivity and supporting precision care across the diagnostic pathway for ATB. To date, however, no scoping review has synthesized AI and ML applications specifically for ATB diagnosis and differential diagnosis while critically appraising reporting quality, risk of bias and explicitly linking to existing clinical frameworks. We therefore conducted this Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews-compliant scoping review[8] to map the existing evidence, categorize applications by research theme and modality, appraise reporting quality, identify key gaps, and propose a translational roadmap for safe and equitable clinical integration that augments rather than replaces current evidence-based pathways.

MATERIALS AND METHODS

This systematic review was conducted according to a pre-defined protocol following Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews guidelines[8]. The review question was framed using the patient-centered care framework[9]: Population - patients with suspected or confirmed ATB; concept - applications of AI or ML; context - diagnosis or differential diagnosis vs Crohn’s disease, peritoneal carcinomatosis, lymphoma, or other mimics in any clinical setting.

Searches were conducted in PubMed [Medical Subject Heading (MeSH)-adapted], Cochrane Library, and Google Scholar. They were executed between April 8, 2026 and 10 April 10, 2026. Only English-language peer-reviewed articles were included, with no lower date limit. These databases were selected for their strengths: PubMed for comprehensive biomedical and clinical literature coverage, Cochrane Library for systematic reviews, and Google Scholar for broad scholarly indexing that includes recent preprints and regional publications.

The full reproducible PubMed search string was as follows: [“Artificial Intelligence” (MeSH) OR “Machine Learning” (MeSH) OR “Deep Learning” (MeSH) OR radiomics (tiab) OR CNN (tiab) OR XGBoost (tiab) OR “few-shot learning” (tiab) OR NLP (tiab)] AND [“Tuberculosis” (MeSH) OR “abdominal tuberculosis” (tiab) OR “intestinal tuberculosis” (tiab) OR “peritoneal tuberculosis” (tiab) OR “tuberculous peritonitis” (tiab) OR “gastrointestinal tuberculosis” (tiab)]. Equivalent adapted search strings were used in the Cochrane Library and Google Scholar.

Inclusion criteria comprised peer-reviewed original research or systematic reviews applying AI or ML to human ATB diagnosis or differential diagnosis. Exclusion criteria comprised exclusive pulmonary tuberculosis focus, no AI or ML component, animal or in vitro studies, editorials, letters, or conference abstracts without full data, and non-peer-reviewed material.

Two independent reviewers performed title/abstract and full-text screening using Rayyan[10]. Inter-rater agreement was substantial (Cohen’s κ = 0.82 for title/abstract screening; κ = 0.78 for full-text screening). Conflicts were resolved through discussion and consensus, with arbitration by the senior author when necessary. Data extraction used a standardized Research Electronic Data Capture form piloted on five studies and completed independently in duplicate[11]. Variables charted included study setting, design, sample size, AI/ML technique, modality, performance metrics [area under the curve (AUC), accuracy, sensitivity, specificity], reference standard, explainability methods, and thematic category.

In addition to performance extraction, we performed a structured appraisal of reporting quality and risk of bias elements aligned with Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (for prediction model studies)[12], Minimum Information about Clinical Artificial Intelligence Modeling (MI-CLAIM) checklist principles[13], and Prediction Model Risk of Bias Assessment Tool domains[14] where applicable. Common deficiencies were summarized narratively. Studies were thematically categorized as intestinal tuberculosis vs Crohn’s disease, peritoneal tuberculosis vs carcinomatosis, nodal tuberculosis vs lymphoma, multimodal integration, or novel/emerging approaches. No patients or members of the public were involved in the design, conduct, or reporting of this scoping review. This is common in early-stage scoping reviews of emerging technologies, where the primary aim is evidence mapping rather than direct intervention co-design[15].

RESULTS

We identified 227 records; after deduplication, 119 unique articles were screened. Seventy-six were excluded at title and abstract stage and 26 of the 43 full texts assessed were excluded (13 not ATB-specific and/or AI/ML applications specific, 6 insufficient data, 4 non-original, 3 pulmonary or unrelated). Seventeen studies met inclusion criteria: 16 original retrospective studies (published 2020-2025, with clear acceleration after 2021) and one international systematic review (Sachan et al[16], 2024) as summarized in Figure 1 and Table 1[16-32]. The Sachan et al's review[16] synthesized evidence on AI for discriminating Crohn disease from gastrointestinal tuberculosis up to early 2024 and highlighted generally good discriminatory performance of radiomics and deep-learning models while noting important limitations in study design and external validation[16].

Figure 1
Figure 1 Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews flow diagram. AI: Artificial intelligence; ML: Machine learning; TB: Tuberculosis; ATB: Abdominal tuberculosis; CT: Computed tomography; NLP: Natural language processing.
Table 1 Characteristics and performance metrics of artificial intelligence and machine learning 16 original studies in abdominal tuberculosis diagnosis.
Ref.
Country
Design and sample
Target
Primary modality
AI/ML technique
Key performance
Explainability
Key notes
Park et al[17], 2025South KoreaRetrospective; internal 1132 CD + 1045 GITB images (130 points + 123 points; train/value/test), external 67 CD + 63 GITB images (17 points + 14 points, no overlap). Colonoscopy (white-light lesion images)ITB vs CD (colonoscopic)Colonoscopy (white-light lesion images)Deep learning/CNN
UNet-ResNet50 CNN (multitask learning)
Internal: Acc 97.7% (AUROC = 0.997). External: Acc 81.5% (AUROC = 0.877)Human crossover trial (6 endoscopists, 780 interpretations): Pooled acc ↑86.2% to 88.8% (P = 0.010), trainees ↑76.7% to 81.0% (P = 0.002)Most rigorous image-based study with external validation + human-AI crossover trial; demonstrates trainee benefit (+4.3% Acc) and addresses prior gaps in raw-image AI, generalizability, and clinical utility evaluation
Cheng et al[18], 2024ChinaRetrospective; 330 patients (value 1 = 224, value 2 = 106); SMOTECD vs ITBCT enterography (arterial + venous phases)DL radiomics + LASSO + logisticArterial-venous DL AUC = 0.885/0.877/0.800 (train/value 1/value 2)None reportedMulti-phase fusion key; DL outperformed handcrafted in arterial phase
Shu et al[19], 2024ChinaRetrospective; 241 patients (51 parameters); real-world validationITB vs CDClinical/EHR/laboratory (51 parameters)XGBoost + SHAP + LIMEAUC = 0.946; Acc 0.884; real-world Acc 0.860, Sens 0.833, Spec 0.871SHAP + LIME; top features T-SPOT, pulmonary TBStrong MDT agreement (90.7%, kappa 0.78); ready for deployment
Lu et al[20], 2024ChinaRetrospective; 82 patients (25 ITB, 57 CD); train 54, test 28ITB vs CDMultimodal (MRE radiomics + colonoscopy radiomics)LASSO + DL + logistic; fused model (33 features)Multidisciplinary AUC = 0.94 (test); single-modality 0.68-0.90None reportedFirst true multidisciplinary fusion; superior to all singles (DeLong P < 0.05)
Liu et al[21], 2024ChinaRetrospective; 85 cases (1973 WSI)CD vs ITB (surgical specimens)Histopathology whole-slide imagesDL CNN; attention mapsCase-level AUC = 0.886 (train)/0.893 (test); slide-level 0.954/0.827Attention maps; top-10 featuresOutperformed juniors; comparable to senior GI pathologists
Li et al[22], 2024ChinaRetrospective; 72 paraffin sectionsCD vs ITBATR-FTIR spectroscopy (mid-IR on paraffin)Spectroscopy + XGBoost/MLHigh accuracy and specificity (detailed in full text)None reportedRapid chemical fingerprinting; non-destructive
Lin et al[23], 2024ChinaRetrospective; 122 endoscopic images (99 CD, 23 ITB)ITB vs CDWhite-light endoscopic (ileum)Few-shot (2-way 3-shot; Xception + dual transfer)AUC = 0.81 (dual transfer); better than single transfer (0.56) & endoscopistsNone reportedHighly data-efficient for rare ITB; dual transfer learning
Shen et al[24], 2024ChinaRetrospective; 301 patients (107 ATB-LN + 194 lymphoma)Nodal TB vs lymphomaCECT (abdominal LNs, 3D VOI)Radiomics (8 features); LR (primary)LR AUC = 0.91 (train)/0.86 (value); Acc 88%/83%; Sens 76%/72%None reportedStrong shape/texture signature; excluded small/necrotic nodes
Pang et al[25], 2023China (multicenter)Retrospective; 178 patients (88 PTB + 90 CD)PTB vs PCCT (omental, peritoneal, ascites, LN signs)ML ensemble on selected featuresAUC = 0.971 (train)/0.914 (test); F1 0.923/0.867None reportedMulticentre; simple interpretable CT signs; strong performance
Gong et al[26], 2023ChinaRetrospective; 105 patients (61 CD, 44 ITB); 5-fold cross-validation (single-center)ITB vs CDCT enterography (multiregional radiomics)Multiregional radiomics model (CT enterography-based)Combined (nomogram) model: AUC = 0.975 (train 95%CI: 0.953–0.998)/0.958 (CV 95%CI: 0.925–0.991); Acc 89.5%; Sens 86.9%; Spec 93.2% (DeLong P =0.004 vs clinical). Radscore (VOI1+VOI2) AUC = 0.962/0.926Nomogram (visual)Multiregional CTE radiomics (intestinal wall + largest LN) + clinical (involved segments) + endoscopy (longitudinal ulcer) via logistic nomogram; highest clinical benefit (DCA); Acc 89.5% superior to readers (66.7%-75.2%); good calibration; single-center Chinese cohort. Completes the 16 studies with visual nomogram for CD vs ITB differentiation
Lu et al[27], 2023ChinaRetrospective; 1271 EHR patients (875 CD, 396 ITB)CD vs ITB (colonoscopy text)Free-text EHR (NLP)Debiased TextCNN (distilled; integrated gradients)Acc 0.83 (noisy value 0.70); superior to non-debiasedIntegrated gradients; 39 phrases (17 guideline-based)Trustworthy AI (accuracy + interpretability + robustness)
Chen et al[28], 2022ChinaRetrospective; 287 cases (59 indicators)ITB vs CDClinical + laboratory + endoscopic (multimodal)Fusion correlation neural networkAverage accuracy > 91%Interpretable rulesPerianal fistulas, comb sign, cobblestone to CD; ascites to ITB
Weng et al[29], 2022ChinaRetrospective; 200 patients (160 CD, 40 ITB)ITB vs CDClinical + radiological (9 variables)XGBoost (best); SHAPAUC = 0.891; Sens 0.813; Spec 0.969; MCC 0.801SHAP (global and local)First SHAP use; perianal fistulas and comb sign point to CD
Zhu et al[30], 2021ChinaRetrospective; 160 patients (93 CD, 67 ITB)CD vs ITB (ileocecal)CT enterography radiomics + clinicalRadiomics (9 GBDT features) + 2 clinicalAUC = 0.96 (train)/0.93 (value); superior to singles (P < 0.01)Nomogram (visual)Excellent bedside translation; helps treatment decisions
Kim et al[31], 2021South KoreaRetrospective; 6617 colonoscopy images (211 CD, 299 BD, 217 ITB)3-class CD vs BD vs ITB (pairwise)Colonoscopy images (white-light)DL CNN; Grad-CAM3-class Acc 65.15% (all)/72.01% (typical); AUROC = 0.78-0.86Grad-CAMBest on typical images; helpful for inexperienced endoscopists
Tong et al[32], 2020ChinaRetrospective; 6,399 patients (5,128 UC, 875 CD, 396 ITB)3-class UC vs CD vs ITB (pairwise)Free-text endoscopic reports (NLP)RF and TextCNNRF: UC-CD 0.89/0.84, UC-ITB 0.83/0.82, CD-ITB 0.72/0.77 (sens/spec)None reportedLargest NLP cohort; excellent for inexperienced clinicians

Our review incorporates five additional original studies published after the Sachan et al's search window[16] and applies a structured appraisal of reporting quality and risk-of-bias elements not previously undertaken (Figure 2). Risk of bias assessment revealed that most studies were at high or unclear overall risk, driven by retrospective single-center designs, composite reference standards, and limited external validation. Only a minority of recent studies incorporated external validation or real-world testing. Geographic origin of original studies (Figure 3) was China (n = 14/16, 87.5%) and South Korea (n = 2, 12.5%). Sample sizes ranged from 66 to 6617 patients or images.

Figure 2
Figure 2 Risk of bias summary plot for the 17 included studies. AI: Artificial intelligence; ML: Machine learning; PROBAST: Prediction model Risk Of Bias ASsessment Tool; QUADAS-2: Quality Assessment of Diagnostic Accuracy Studies-2; ROB: Risk of bias; ITB: Intestinal tuberculosis.
Figure 3
Figure 3 Geographic distribution of the 16 original studies. China: 14 (87.5%); Korea: 2. All 16 original retrospective studies originated exclusively from East Asia. No studies from any other country or region were identified in the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews scoping review (search completed April 2026).

Of the 16 original studies, 14 focused primarily on differentiating intestinal tuberculosis from Crohn’s disease, while one each addressed peritoneal tuberculosis vs carcinomatosis and nodal tuberculosis vs lymphoma (Table 2). Four studies incorporated multimodal integration, and several explored emerging techniques, including few-shot learning, spectroscopy-assisted ML, debiased natural language processing, and explainable AI approaches. Despite explicit inclusion criteria encompassing management, no studies applied AI or ML to treatment response prediction, surgical decision support, prognosis, or drug-resistance forecasting in ATB. HIV status was reported in fewer than 20% of studies, with no stratification by immunosuppression or drug-resistant tuberculosis status. Reference standards were composite clinical criteria in most studies; only 4 of 16 original studies used microbiological confirmation as the primary reference standard. This contrasts with the STW’s preference for microbiological positivity (culture/nucleic acid amplification test) or definitive histological features (caseating granulomas) where feasible, and its use of composite criteria plus ascitic adenosine deaminase only when microbiology is negative[6]. No study addressed solid-organ visceral forms of ATB in depth.

Table 2 Categorization of artificial intelligence and machine learning techniques according to research theme and representative studies in abdominal tuberculosis.
Theme
Imaging/point data
Clinical/areal data
Key techniques
Representative original studies (n = 16)
Intestinal tuberculosis vs Crohn’s diseaseEndoscopy images (white-light colonoscopy), CT/MRE radiomics, histopathology WSI, ATR-FTIR spectroscopyClinical/EHR parameters, nomograms, T-SPOT, pulmonary TB historyCNN/deep learning (U-Net, ResNet), radiomics (handcrafted + deep learning), XGBoost, few-shot learning (Xception), NLP/TextCNN (debiased), multimodal fusion, SHAP/LIME/attention maps explainabilityPark et al[17], 2025; Cheng et al[18], 2024; Shu et al[19], 2024; Liu et al[21], 2024; Li et al[22], 2024; Lin et al[23], 2024; Gong et al[26], 2023; Lu et al[27], 2023; Chen et al[28], 2022; Weng et al[29], 2022; Zhu et al[30], 2021; Kim et al[31], 2021; Tong et al[32], 2020
Peritoneal tuberculosis vs carcinomatosisCT (omental/peritoneal signs, ascites, lymph nodes, scalloping)Omental/peritoneal imaging signsMachine learning ensembles, radiomics (shape, first-order, texture features)Pang et al[25], 2023
Nodal tuberculosis vs lymphomaContrast-enhanced CT (lymph node features)Lymph-node morphological/textural featuresRadiomics (logistic regression/SVM)Shen et al[24], 2024
Multimodal integrationFusion of MRE radiomics + colonoscopy images + histopathology WSIIntegrated clinical + endoscopic + radiological + pathological featuresMultidisciplinary fusion models (LASSO + logistic regression), fusion correlation neural networksLu et al[20], 2024; Chen et al[28], 2022
Novel/emerging approachesFew-shot endoscopic images, ATR-FTIR spectroscopy, noisy EHR textClinical parameters (for few-shot/explainable models)Few-shot learning (dual transfer), spectroscopy + XGBoost/ML, debiased TextCNN (NLP), explainable AI (SHAP, LIME, integrated gradients, attention maps)Lin et al[23], 2024; Li et al[22], 2024; Lu et al[27], 2023 (also embedded in multiple intestinal TB vs CD studies above)
Granular performance synthesis

Overall performance reached AUC = 0.80-0.96 and accuracy 69%-100% in optimized models. Multimodal and explainable models consistently outperformed single-modality approaches. Temporal publication trends (2020-2025) and modality distribution is shown in Figure 4.

Figure 4
Figure 4 Temporal publication trends (2020-2025) and modality distribution. Updated with n = 16 original articles (+ 1 systematic review = 17 total). Sixteen original retrospective studies (total n = 16). Excludes the systematic review by Sachan et al[16]. CT: Computed tomography; MRE: Magnetic resonance elastography; CNN: Convolutional neural network; NLP: Natural language processing.

Radiomics (computed tomography/magnetic resonance elastography): Strong performance across studies. Zhu et al[30] (2021) achieved AUC = 0.96 (train)/0.93 (validation) using computed tomography enterography radiomics plus clinical variables with a visual nomogram. Cheng et al[18] (2024) reported arterial-venous deep learning radiomics AUCs of approximately 0.88. Multiregional radiomics approaches further improved discrimination.

Endoscopic deep learning/convolutional neural network: High internal accuracy with emerging external validation. Park et al[17] (2025) demonstrated a multitask UNet-ResNet50 convolutional neural network (CNN) on white-light colonoscopy images with internal accuracy 97.7% [area under the receiver operating characteristic curve (AUROC) = 0.997] and external accuracy 81.5% (AUROC = 0.877). A human crossover trial with six endoscopists showed pooled accuracy improvement from 86.2% to 88.8% (P = 0.010), with greater benefit for trainees.

Multimodal and explainable AI: Consistently superior. Lu et al[20] (2024) multidisciplinary fusion model (magnetic resonance elastography radiomics + colonoscopy radiomics) achieved AUC = 0.94 on the test set, significantly outperforming single-modality models. Shu et al’s XG Boost model (2024)[19] on 51 clinical/electronic health record parameters reached AUROC = 0.946 (accuracy 0.884), with real-world validation accuracy 0.860, sensitivity 0.833, and specificity 0.871; strong agreement with multidisciplinary teams (90.7%, κ = 0.78) and interpretable via SHapley Additive exPlanations/local interpretable model-agnostic explanations (top features: T-SPOT, pulmonary tuberculosis history).

Histopathology whole-slide imaging: Liu et al's deep learning model (2024)[21] on 1973 whole-slide images achieved case-level AUC = 0.886 (internal)/0.893 (external) and outperformed junior pathologists while matching experienced gastrointestinal pathologists. Attention maps provided interpretability.

Emerging approaches: Few-shot learning (Lin et al[23], 2024) offered data-efficient classification for rare intestinal tuberculosis cases. attenuated total reflection-Fourier transform infrared spectroscopy spectroscopy combined with ML (Li et al[22], 2024) provided rapid, non-destructive chemical fingerprinting. Debiased Text CNN (Lu et al[27], 2023) improved robustness on noisy electronic health record text.

Performance was highest in multimodal and explainable models. Standout examples include Park et al[17] (2025) endoscopic multitask CNN with external validation and demonstrated improvement in endoscopist accuracy, and Liu et al[21] (2024) whole-slide imaging deep learning that matched expert gastrointestinal pathologists. These findings are consistent with the performance range summarized by Sachan et al[16] (2024), although multimodal and explainable approaches have shown further gains in more recent work.

Critical appraisal of reference standards and equity gaps

Most studies relied on composite clinical criteria rather than microbiological confirmation as the primary reference standard. This raises the risk that models may replicate clinician diagnostic bias rather than true biological distinctions, particularly when aligned against hierarchical criteria that prioritize microbiological or caseating granuloma evidence[32,33]. No studies stratified performance by HIV status, immunosuppression, or drug-resistant tuberculosis. Solid-organ visceral ATB forms remained largely unaddressed. These gaps directly inform the priorities of the translational roadmap below.

Sachan et al[16] (2024) reached broadly similar conclusions regarding the promise of radiomics and endoscopic deep learning for intestinal tuberculosis vs Crohn disease differentiation. However, their review did not assess reporting quality against contemporary standards (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis, MI-CLAIM, Prediction Model Risk of Bias Assessment Tool), did not examine peritoneal or nodal tuberculosis applications in detail, and did not link findings to operational clinical pathways such as India’s ICMR STW[6].

The overwhelming predominance of East Asian cohorts (100% of the 16 original studies) means that the synthesized performance metrics and identified best practices primarily reflect populations with relatively lower HIV co-infection rates and potentially different mycobacterial/host factors compared with South Asian, African, or Latin American high-burden settings. This geographic concentration limits confident extrapolation and reinforces the roadmap’s call for urgent prospective validation in diverse LMIC populations. Small ATB subsample sizes in several studies further elevate overfitting risk, as already flagged in the risk-of-bias appraisal. The current scoping review therefore provides both updated evidence maps and a more granular appraisal of methodological and equity considerations.

DISCUSSION

This scoping review of 17 studies demonstrates that AI and ML can substantially enhance the differentiation of ATB from its major mimics. Radiomics based on computed tomography or magnetic resonance enterography, together with endoscopic deep learning using CNNs, remain the dominant approaches. However, multimodal fusion models consistently achieve the highest performance. Novel techniques, including few-shot learning for data-scarce settings, spectroscopy-assisted ML, debiased natural language processing, and explainable AI, directly address practical barriers to adoption in LMIC.

Contextualization with conventional diagnostics

AI and ML are particularly valuable because conventional tests have well-documented limitations. Xpert Mycobacterium tuberculosis/rifampicin sensitivity remains low on intestinal tissue (approximately 23% vs composite reference) and ascitic fluid (approximately 30%-64% vs culture)[34]. Culture yields are modest (approximately 35%-40%)[35]. Histological features such as caseating granulomas offer high specificity but low sensitivity (approximately 21%), while confluent granulomas and histiocyte-lined ulcers provide supportive but imperfect discrimination from Crohn’s disease[36]. Imaging signs show significant overlap. These constraints, emphasized in evidence-based hierarchical frameworks[5], create precisely the subjectivity and diagnostic uncertainty that well-designed AI tools can mitigate.

Bias, transportability, and equity considerations

Overwhelmingly retrospective, single-center designs (predominantly East Asian cohorts) introduce spectrum bias and raise concerns about overfitting with small intestinal tuberculosis samples. Transportability of China-trained models to Indian, African, and other populations remains untested. These populations differ in HIV prevalence, malnutrition rates, and potentially mycobacterial strain characteristics, raising substantive equity concerns[37]. Infrastructure gaps in remote LMIC clinics (reliable internet, electronic health record linkage, offline capability) are unresolved[38]. Tuberculosis-related stigma and patient privacy concerns require explicit attention in deployment frameworks[39].

Potential harms and unintended consequences

While AI and ML tools can improve ATB differentiation, they carry several risks. False-negative predictions may delay anti-tubercular therapy in true cases, allowing disease progression and continued transmission[34]. False-positive outputs risk exposing patients with Crohn’s disease or other mimics to unnecessary anti-tuberculosis treatment or corticosteroids[36]. Overreliance on algorithmic outputs, especially among trainees, may erode clinical judgment and diagnostic skills when AI recommendations conflict with nuanced findings or hierarchical criteria[5]. Automated labeling also risks amplifying tuberculosis-related stigma[39]. These harms can be mitigated through mandatory human-in-the-loop design, built-in explainability features, prospective safety monitoring (including adverse-event tracking and subgroup analyses) embedded in phase 2 and phase 3 studies of the translational roadmap, and explicit alignment of AI outputs with ICMR STW decision nodes so that tools support rather than supplant established evidence-based pathways[6,14].

Linkage to existing guidelines and comparison with pulmonary applications

AI and ML should be positioned as adjuncts that augment, rather than replace, established hierarchical diagnostic criteria and objective response assessment (early mucosal healing at two months, ascites resolution, fecal calprotectin trends)[40]. Pulmonary tuberculosis AI applications (particularly chest X-ray computer-aided detection) are more mature[41]; abdominal applications lag due to greater data scarcity, the complexity of differentiation tasks, and smaller event rates for intestinal tuberculosis. This review therefore provides both a map of current capabilities, based on 16 original studies summarized in Table 1, and a clear agenda for achieving clinically meaningful impact.

Both this review and the earlier systematic review by Sachan et al[16] (2024) underscore that AI applications for ATB remain at an early stage. In contrast, computer-aided detection systems for pulmonary tuberculosis are considerably more mature. While Sachan et al[16] focused on diagnostic discrimination performance, the present work additionally emphasizes the need to align future AI tools with hierarchical diagnostic criteria and response-assessment timelines already embedded in national guidelines.

Future directions and translational roadmap

AI and ML offer the potential to improve diagnostic precision in ATB and support timely initiation of anti-tubercular therapy. However, these technologies are best positioned as adjuncts that augment clinical judgment, microbiological testing, radiological expertise, and histopathology, rather than as replacements[42]. ML-based predictive models can estimate the probability of disease and thereby reduce diagnostic delays, while AI-assisted triage systems may help prioritize high-risk patients for early investigations in resource-constrained settings[43]. In addition, the ability of these tools to analyze large datasets presents opportunities for the discovery of novel biomarkers and phenotypic subtypes of ATB[44].

To translate promising retrospective models into equitable clinical impact, we propose a structured four-phase translational roadmap with defined timelines, success metrics, and multi-stakeholder engagement as illustrated in Figure 5.

Figure 5
Figure 5 Four-phase translational roadmap for artificial intelligence and machine learning applications in abdominal tuberculosis. This roadmap prioritizes microbiological reference standards, low-income and middle-income countries leadership, equity, and integration with existing national frameworks (e.g., Indian Council of Medical Research, standard treatment workflow). MI-CLAIM: Minimum Information about Clinical Artificial Intelligence Modeling; HIV: Human immunodeficiency virus; LMIC: Low-income and middle-income countries; AI: Artificial intelligence; CDSCO: Central Drugs Standard Control Organization; FDA: Food and Drug Administration; CE: Conformité Européenne; STW: Standard treatment workflow; NTEP: National Tuberculosis Elimination Programme; WHO: World Health Organization.

Phase 1 foundation (2020-2026): This phase involves retrospective model development, internal validation, open dataset creation, and standardized reporting in accordance with the MI-CLAIM checklist[16]. To strengthen the evidence base for future research, we recommend establishing a minimal core dataset for all ATB-AI studies. This should include demographics, HIV status, immunosuppression, drug-resistance markers, alignment with the Jha et al[5] hierarchical reference standard, and standardized imaging and endoscopy protocols. In addition, studies should incorporate a microbiological confirmation subset (target ≥ 30% of intestinal tuberculosis cases) for model training and validation, while systematically capturing variables defined in the ICMR STW, including ascitic adenosine deaminase levels, histological features, treatment response at 4 weeks and 8 weeks, and National Tuberculosis Elimination Program (NTEP) treatment categories[6]. This 30% microbiological confirmation target (culture, nucleic acid amplification test/Xpert, or definitive caseating granulomas) is proposed as a pragmatic and feasible benchmark informed by diagnostic yield data from ATB cohorts in endemic LMIC settings. Reported combined microbiological positivity rates in intestinal tissue or ascitic fluid range from approximately 24% to over 50% in optimized prospective series from high-burden regions, depending on sampling adequacy and test combinations. A 30% subset is therefore realistic for well-designed multicenter studies employing standardized protocols (multiple targeted endoscopic biopsies plus laparoscopic sampling when indicated) while providing a meaningful anchor for model training and validation against biological rather than purely composite reference standards, consistent with the hierarchical criteria of the ICMR STW[6] and Jha et al[5].

Phase 2 prospective validation (2026-2028): Phase 2 focuses on generating robust, generalizable evidence through multicenter prospective cohorts across diverse LMIC settings. We recommend establishing at least six sites (India, South Africa, Indonesia, Nigeria, Brazil, and Pakistan), with a minimum of 400 participants per site and adequate power to evaluate performance in HIV-positive and drug-resistant tuberculosis subgroups. Studies should incorporate real-world variability, rigorous external validation, head-to-head comparison against current Jha et al’s criteria[5] and ICMR STW pathways[6], and prospective assessment of diagnostic yield. The primary endpoint will be diagnostic accuracy (AUC, sensitivity, and specificity). Secondary endpoints include reduction in diagnostic delay, rates of unnecessary laparoscopy, and cost per correct diagnosis. Federated learning combined with systematic bias audits assessing demographic parity across ethnicity and HIV status is strongly encouraged[45]. Success will be defined as achieving an external validation AUC ≥ 0.85 in diverse LMIC cohorts with a clinically meaningful reduction in diagnostic delay.

Phase 3 clinical integration (2028-2031): This phase centers on the development and deployment of practical, regulatory-approved AI tools designed to integrate seamlessly into existing national diagnostic workflows. Priority tools include real-time endoscopic AI systems, multimodal decision-support applications, explainable radiomics nomograms, and treatment-response predictors[43]. These should support key decision nodes within the ICMR STW[6], such as real-time assistance when diagnosis remains uncertain, prediction of treatment response at 4-week and 8-week assessments, and automated flagging for surgical or endoscopic intervention. Regulatory approval pathways should be pursued through the Central Drugs Standard Control Organization (India), the United States Food and Drug Administration, and Conformité Européenne marking for software as a medical device[46]. Implementation must include structured clinician training, seamless electronic health record integration, and robust consent and privacy frameworks that address tuberculosis-related stigma[39]. Success metrics include at least 80% clinician adoption in pilot sites within 12 months and a minimum 50% reduction in inappropriate anti-tuberculosis therapy or steroid exposure among patients ultimately diagnosed with Crohn’s disease or other mimics. Economic evaluations embedded in phase 3 will model modality-specific cost drivers (imaging and endoscopy AI infrastructure vs downstream savings from reduced unnecessary laparoscopy and surgery, historically occurring in approximately 15% of cases) alongside diagnostic accuracy gains[5]. Key partners should include National Tuberculosis Programs, the Stop Tuberculosis Partnership, the foundation for innovative new diagnostics, and patient advocacy groups.

Phase 4 implementation and equity (2031 and beyond): The final phase emphasizes large-scale, sustainable implementation with equity as a central principle. Rigorous cost-effectiveness analyses and implementation research should evaluate the impact of AI tools on key clinical and health system outcomes, including diagnostic delay, unnecessary surgery, overuse of anti-tuberculosis therapy, and mortality. Equity considerations must be embedded throughout, including the development of edge-computing and offline-capable solutions suitable for resource-constrained settings, systematic bias audits using metrics such as demographic parity, and data-privacy frameworks that explicitly address tuberculosis stigma[39]. Federated learning approaches should be adopted to enable secure multi-center collaboration[45]. Close engagement with National Tuberculosis Programs, particularly India’s NTEP, alongside patient advocacy groups, the World Health Organization Global Tuberculosis Program, and the Stop Tuberculosis Partnership will be essential to support large-scale implementation[47]. To support this agenda, targeted funding should be sought from the Bill & Melinda Gates Foundation, the Wellcome Trust, and national research councils to establish LMIC-led research consortia[48] with explicit deliverables related to equity, infrastructure development, and measurable reductions in inappropriate anti-tuberculosis therapy use among patients with conditions that mimic Crohn’s disease.

This roadmap provides a pragmatic, equity-centered pathway from current retrospective evidence to real-world deployment, with the explicit goal of complementing and strengthening existing evidence-based frameworks rather than creating parallel systems. Feasibility of the proposed phases will be enhanced by alignment with established governance frameworks of National Tuberculosis Programs (e.g., India’s NTEP), ethics harmonization via national research councils, and data-sharing and capacity-building mechanisms facilitated by the foundation for innovative new diagnostics and the Stop Tuberculosis Partnership[6,47].

Limitations

Most included studies did not report detailed quantitative imaging features (such as bowel wall thickness, number of necrotic lymph nodes, or extent of ileocecal and peritoneal involvement) or correlate these with AI model performance. Furthermore, large multicontinental, prospectively collected datasets with microbiological confirmation are urgently needed before robust predictive models can be developed and integrated with clinical, biomarker, and histopathological data. Although Google Scholar was included to broaden coverage, the restriction to English-language publications may have missed relevant studies published in regional journals or non-English databases (e.g., China National Knowledge Infrastructure, Latin American and Caribbean Health Sciences Literature). Future updates of this review should incorporate these sources to better reflect the full global evidence base.

CONCLUSION

AI and ML offer a powerful, objective adjunct for ATB diagnosis and differential diagnosis. By systematically addressing the identified gaps through the proposed translational roadmap, these technologies can reduce diagnostic uncertainty and minimize inappropriate therapy and surgery. The emphasis must be on prospective, diverse, explainable, and implementation-focused research, including the urgent development of management applications. This will advance equitable precision care in high-burden settings worldwide by complementing and enhancing practical national frameworks such as India’s ICMR STW for Adult ATB. Future work must prioritize real-world validation, global collaboration led by researchers from LMICs, explicit development of management-focused AI applications, emerging foundation models, and infrastructure-appropriate deployment strategies.

References
1.  Tobin EH, Khatri AM.   Abdominal Tuberculosis. 2025 Feb 6. In: StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing; 2026 Jan-.  [PubMed]  [DOI]
2.  World Health Organization  Global tuberculosis report 2025. Geneva: World Health Organization, 2025.  [PubMed]  [DOI]
3.  Press Information Bureau, Delhi.   TB incidence in India drops by 21% from 237 per lakh population in 2015 to 187 per lakh population in 2024. The NewsHour. 12 Nov 2025. Available from: https://www.thenewshour.org/other-news/tb-incidence-in-india-drops-by-21-from-237-per-lakh-population-in-2015-to-187-per-lakh-population-in-2024-almost-double-the-rate-of-decline-observed-globally/.  [PubMed]  [DOI]
4.  Al-Zanbagi AB, Shariff MK. Gastrointestinal tuberculosis: A systematic review of epidemiology, presentation, diagnosis and treatment. Saudi J Gastroenterol. 2021;27:261-274.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 84]  [Cited by in RCA: 66]  [Article Influence: 13.2]  [Reference Citation Analysis (0)]
5.  Jha DK, Pathiyil MM, Sharma V. Evidence-based approach to diagnosis and management of abdominal tuberculosis. Indian J Gastroenterol. 2023;42:17-31.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 54]  [Cited by in RCA: 54]  [Article Influence: 18.0]  [Reference Citation Analysis (1)]
6.  Standard treatment workflow (STW) for the management of adult abdominal tuberculosis.  [Internet]. [cited 15 March 2026]. Available from: https://www.icmr.gov.in/icmrobject/uploads/STWs/1725964690_1_adult_abdominal_tb_18032022.pdf.  [PubMed]  [DOI]
7.  Qin ZZ, Sander MS, Rai B, Titahong CN, Sudrungrot S, Laah SN, Adhikari LM, Carter EJ, Puri L, Codlin AJ, Creswell J. Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems. Sci Rep. 2019;9:15000.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 136]  [Cited by in RCA: 181]  [Article Influence: 25.9]  [Reference Citation Analysis (0)]
8.  Tricco AC, Lillie E, Zarin W, O'Brien KK, Colquhoun H, Levac D, Moher D, Peters MDJ, Horsley T, Weeks L, Hempel S, Akl EA, Chang C, McGowan J, Stewart L, Hartling L, Aldcroft A, Wilson MG, Garritty C, Lewin S, Godfrey CM, Macdonald MT, Langlois EV, Soares-Weiser K, Moriarty J, Clifford T, Tunçalp Ö, Straus SE. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med. 2018;169:467-473.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 31615]  [Cited by in RCA: 24638]  [Article Influence: 3079.8]  [Reference Citation Analysis (5)]
9.  Peters MD, Godfrey CM, Khalil H, McInerney P, Parker D, Soares CB. Guidance for conducting systematic scoping reviews. Int J Evid Based Healthc. 2015;13:141-146.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 5522]  [Cited by in RCA: 4121]  [Article Influence: 374.6]  [Reference Citation Analysis (3)]
10.  Ouzzani M, Hammady H, Fedorowicz Z, Elmagarmid A. Rayyan-a web and mobile app for systematic reviews. Syst Rev. 2016;5:210.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 18843]  [Cited by in RCA: 15872]  [Article Influence: 1587.2]  [Reference Citation Analysis (7)]
11.  Harris PA, Taylor R, Minor BL, Elliott V, Fernandez M, O'Neal L, McLeod L, Delacqua G, Delacqua F, Kirby J, Duda SN; REDCap Consortium. The REDCap consortium: Building an international community of software platform partners. J Biomed Inform. 2019;95:103208.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 20177]  [Cited by in RCA: 18452]  [Article Influence: 2636.0]  [Reference Citation Analysis (7)]
12.  Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med. 2015;162:55-63.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 2350]  [Cited by in RCA: 2166]  [Article Influence: 196.9]  [Reference Citation Analysis (3)]
13.  Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, Reitsma JB, Kleijnen J, Mallett S; PROBAST Group†. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med. 2019;170:51-58.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 2040]  [Cited by in RCA: 1810]  [Article Influence: 258.6]  [Reference Citation Analysis (5)]
14.  Norgeot B, Quer G, Beaulieu-Jones BK, Torkamani A, Dias R, Gianfrancesco M, Arnaout R, Kohane IS, Saria S, Topol E, Obermeyer Z, Yu B, Butte AJ. Minimum information about clinical artificial intelligence modeling: the MI-CLAIM checklist. Nat Med. 2020;26:1320-1324.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 215]  [Cited by in RCA: 388]  [Article Influence: 64.7]  [Reference Citation Analysis (5)]
15.  Staniszewska S, Brett J, Simera I, Seers K, Mockford C, Goodlad S, Altman DG, Moher D, Barber R, Denegri S, Entwistle A, Littlejohns P, Morris C, Suleman R, Thomas V, Tysall C. GRIPP2 reporting checklists: tools to improve reporting of patient and public involvement in research. Res Involv Engagem. 2017;3:13.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 289]  [Cited by in RCA: 427]  [Article Influence: 47.4]  [Reference Citation Analysis (0)]
16.  Sachan A, Kakadiya R, Mishra S, Kumar-M P, Jena A, Gupta P, Sebastian S, Deepak P, Sharma V. Artificial intelligence for discrimination of Crohn's disease and gastrointestinal tuberculosis: A systematic review. J Gastroenterol Hepatol. 2024;39:422-430.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 8]  [Cited by in RCA: 12]  [Article Influence: 6.0]  [Reference Citation Analysis (0)]
17.  Park K, Lim J, Shin SH, Ryu M, Shin H, Lee M, Hong SW, Hwang SW, Park SH, Yang DH, Ye BD, Myung SJ, Yang SK, Kim N, Byeon JS. Artificial intelligence-aided colonoscopic differential diagnosis between Crohn's disease and gastrointestinal tuberculosis. J Gastroenterol Hepatol. 2025;40:115-122.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 1]  [Cited by in RCA: 4]  [Article Influence: 4.0]  [Reference Citation Analysis (0)]
18.  Cheng M, Zhang H, Huang W, Li F, Gao J. Deep Learning Radiomics Analysis of CT Imaging for Differentiating Between Crohn's Disease and Intestinal Tuberculosis. J Imaging Inform Med. 2024;37:1516-1528.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 3]  [Cited by in RCA: 8]  [Article Influence: 4.0]  [Reference Citation Analysis (0)]
19.  Shu Y, Chen Z, Chi J, Cheng S, Li H, Liu P, Luo J. A Machine Learning Method for Differentiation Crohn's Disease and Intestinal Tuberculosis. J Multidiscip Healthc. 2024;17:3835-3847.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in RCA: 3]  [Reference Citation Analysis (0)]
20.  Lu B, Huang Z, Lin J, Zhang R, Shen X, Huang L, Wang X, He W, Huang Q, Fang J, Mao R, Li Z, Huang B, Feng ST, Ye Z, Zhang J, Wang Y. A novel multidisciplinary machine learning approach based on clinical, imaging, colonoscopy, and pathology features for distinguishing intestinal tuberculosis from Crohn disease. Abdom Radiol. 2024;49:2187-2197.  [PubMed]  [DOI]  [Full Text]
21.  Liu X, Li F, Xu J, Ma J, Duan X, Mao R, Chen M, Chen Z, Huang Y, Jiang J, Huang BB Ye Z. Deep learning model to differentiate Crohn disease from intestinal tuberculosis using histopathological whole slide images from intestinal specimens. Virchows Arch. 2024;484:965-976.  [PubMed]  [DOI]  [Full Text]
22.  Li YP, Lu TY, Huang FR, Zhang WM, Chen ZQ, Guang PW, Deng LY, Yang XH. Differential diagnosis of Crohn's disease and intestinal tuberculosis based on ATR-FTIR spectroscopy combined with machine learning. World J Gastroenterol. 2024;30:1377-1392.  [PubMed]  [DOI]  [Full Text]
23.  Lin J, Zhu S, Yin M, Xue H, Liu L, Liu X, Liu L, Xu C, Zhu J. Few-shot learning for the classification of intestinal tuberculosis and Crohn's disease on endoscopic images: A novel learn-to-learn framework. Heliyon. 2024;10:e26559.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 5]  [Reference Citation Analysis (0)]
24.  Shen MT, Liu X, Gao Y, Shi R, Jiang L, Yao J. Radiomics-based quantitative contrast-enhanced CT analysis of abdominal lymphadenopathy to differentiate tuberculosis from lymphoma. Precis Clin Med. 2024;7:pbae002.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in RCA: 2]  [Reference Citation Analysis (0)]
25.  Pang Y, Li Y, Xu D, Sun X, Hou D. Differentiating peritoneal tuberculosis and peritoneal carcinomatosis based on a machine learning model with CT: a multicentre study. Abdom Radiol (NY). 2023;48:1545-1553.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 13]  [Reference Citation Analysis (0)]
26.  Gong T, Li M, Pu H, Yin LL, Peng SK, Zhou Z, Zhou M, Li H. Computed tomography enterography-based multiregional radiomics model for differential diagnosis of Crohn's disease from intestinal tuberculosis. Abdom Radiol (NY). 2023;48:1900-1910.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 12]  [Reference Citation Analysis (0)]
27.  Lu K, Tong Y, Yu S, Lin Y, Yang Y, Xu H, Li Y, Yu S. Building a trustworthy AI differential diagnosis application for Crohn's disease and intestinal tuberculosis. BMC Med Inform Decis Mak. 2023;23:160.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 14]  [Reference Citation Analysis (0)]
28.  Chen Y, Li Y, Wu M, Lu F, Hou M, Yin Y. Differentiating Crohn’s disease from intestinal tuberculosis using a fusion correlation neural network. Knowl-Based Syst. 2022;244:108570.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 2]  [Cited by in RCA: 9]  [Article Influence: 2.3]  [Reference Citation Analysis (0)]
29.  Weng F, Meng Y, Lu F, Wang Y, Wang W, Xu L, Cheng D, Zhu J. Differentiation of intestinal tuberculosis and Crohn disease through an explainable machine learning method. Sci Rep. 2022;12:13774.  [PubMed]  [DOI]  [Full Text]
30.  Zhu C, Yu Y, Wang S, Wang X, Gao Y, Li C, Li J, Ge Y, Wu X. A Novel Clinical Radiomics Nomogram to Identify Crohn's Disease from Intestinal Tuberculosis. J Inflamm Res. 2021;14:6511-6521.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 3]  [Cited by in RCA: 28]  [Article Influence: 5.6]  [Reference Citation Analysis (0)]
31.  Kim JM, Kang JG, Kim S, Cheon JH. Deep-learning system for real-time differentiation between Crohn's disease, intestinal Behçet's disease, and intestinal tuberculosis. J Gastroenterol Hepatol. 2021;36:2141-2148.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 6]  [Cited by in RCA: 33]  [Article Influence: 6.6]  [Reference Citation Analysis (0)]
32.  Tong Y, Lu K, Yang Y, Li J, Lin Y, Wu D, Yang A, Li Y, Yu S, Qian J. Can natural language processing help differentiate inflammatory intestinal diseases in China? Models applying random forest and convolutional neural network approaches. BMC Med Inform Decis Mak. 2020;20:248.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 40]  [Cited by in RCA: 33]  [Article Influence: 5.5]  [Reference Citation Analysis (0)]
33.  Jayakumar S, Sounderajah V, Normahani P, Harling L, Markar SR, Ashrafian H, Darzi A. Quality assessment standards in artificial intelligence diagnostic accuracy systematic reviews: a meta-research study. NPJ Digit Med. 2022;5:11.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 42]  [Cited by in RCA: 73]  [Article Influence: 18.3]  [Reference Citation Analysis (0)]
34.  Sharma V, Soni H, Kumar-M P, Dawra S, Mishra S, Mandavdhare HS, Singh H, Dutta U. Diagnostic accuracy of the Xpert MTB/RIF assay for abdominal tuberculosis: a systematic review and meta-analysis. Expert Rev Anti Infect Ther. 2021;19:253-265.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 48]  [Cited by in RCA: 31]  [Article Influence: 6.2]  [Reference Citation Analysis (0)]
35.  Sanai FM, Bzeizi KI. Systematic review: tuberculous peritonitis--presenting features, diagnostic strategies and treatment. Aliment Pharmacol Ther. 2005;22:685-700.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 312]  [Cited by in RCA: 243]  [Article Influence: 11.6]  [Reference Citation Analysis (1)]
36.  Du J, Ma YY, Xiang H, Li YM. Confluent granulomas and ulcers lined by epithelioid histiocytes: new ideal method for differentiation of ITB and CD? A meta analysis. PLoS One. 2014;9:e103303.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 32]  [Cited by in RCA: 33]  [Article Influence: 2.8]  [Reference Citation Analysis (0)]
37.  Musa A, Prasad R, Hernandez M. Addressing cross-population domain shift in chest X-ray classification through supervised adversarial domain adaptation. Sci Rep. 2025;15:11383.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in RCA: 10]  [Reference Citation Analysis (0)]
38.  Al-Ganad A, Al-Shahdhi A, Al-Dhaifi O, Hajeb E, Hajeb H, Al-Motarreb A. Deploying medical AI in low-resource settings: a scoping review of challenges and strategies. Front Digit Health. 2026;8:1743634.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in RCA: 1]  [Reference Citation Analysis (0)]
39.  Nasir M, Ahmed S, Shameem M, Siddiqui K, Karim A. Artificial intelligence driven healthcare and stigma in tuberculosis: Through the lens of ethics. Indian J Tuberc. 2026;73:125-130.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 1]  [Reference Citation Analysis (0)]
40.  Sharma A, Goyal A, Kandasamy D, Kedia S, Ahuja V, Sharma R. Imaging in Abdominal Tuberculosis. Indographics. 2024;3:45-63.  [PubMed]  [DOI]  [Full Text]
41.  Han ZL, Zhang YY, Li J, Gao S, Liu W, Yang WJ, Xing ZH. A systematic review and meta-analysis of artificial intelligence software for tuberculosis diagnosis using chest X-ray imaging. J Thorac Dis. 2025;17:3223-3237.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in RCA: 17]  [Reference Citation Analysis (0)]
42.  Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25:44-56.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 6739]  [Cited by in RCA: 4246]  [Article Influence: 606.6]  [Reference Citation Analysis (9)]
43.  World Health Organization  Ethics and governance of artificial intelligence for health. Geneva: World Health Organization, 2021.  [PubMed]  [DOI]
44.  Rajkomar A, Dean J, Kohane I. Machine Learning in Medicine. N Engl J Med. 2019;380:1347-1358.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in Crossref: 3295]  [Cited by in RCA: 2072]  [Article Influence: 296.0]  [Reference Citation Analysis (4)]
45.  Li S, Wu Q, Li X, Miao D, Hong C, Gu W, Ning Y, Shang Y, Liu N. FairFML: A Unified Approach to Algorithmic Fair Federated Learning with Applications to Reducing Gender Disparities in Cardiac Arrest Outcomes. Stud Health Technol Inform. 2025;329:1848-1849.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Cited by in RCA: 2]  [Reference Citation Analysis (0)]
46.  Chauhan SB, Gaur R, Akram A, Singh I. Artificial Intelligence Driven insights for Regulatory Intelligence in Medical Devices: Evaluating EMA, FDA and CDSCO Frameworks. Glob Clin Eng J. 2025;7:11-24.  [PubMed]  [DOI]  [Full Text]
47.  Ciecierski-Holmes T, Singh R, Axt M, Brenner S, Barteit S. Artificial intelligence for strengthening healthcare systems in low- and middle-income countries: a systematic scoping review. NPJ Digit Med. 2022;5:162.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 73]  [Cited by in RCA: 153]  [Article Influence: 38.3]  [Reference Citation Analysis (0)]
48.  Abimbola S. The foreign gaze: authorship in academic global health. BMJ Glob Health. 2019;4:e002068.  [RCA]  [PubMed]  [DOI]  [Full Text]  [Full Text (PDF)]  [Cited by in Crossref: 179]  [Cited by in RCA: 266]  [Article Influence: 38.0]  [Reference Citation Analysis (0)]
Footnotes

Peer review: Externally peer reviewed.

Peer-review model: Single blind

Corresponding Author's Membership in Professional Societies: Indian Society of Anesthesiologists.

Specialty type: Gastroenterology and hepatology

Country of origin: India

Peer-review report’s classification

Scientific quality: Grade A, Grade B, Grade B

Novelty: Grade A, Grade A, Grade B

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

Scientific significance: Grade A, Grade B, Grade B

P-Reviewer: Banu MRA-, Assistant Professor, India; Vagholkar K, FACS, FRCS (Gen Surg), Full Professor, Visiting Professor, India S-Editor: Zuo Q L-Editor: A P-Editor: Wang WB

Write to the Help Desk