Translational gastroenterology and hepatology · 2026
Artificial intelligence-augmented small bowel capsule endoscopy for coeliac disease: a literature review on accuracy, workflow, and safety.
СтатьяОбзорНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Dhali Arkadeep, Maity Rick, Biswas Jyotirmoy, Hann Alexander, Sidhu Reena, Sanders David Surendran
Аннотация
BACKGROUND AND OBJECTIVE: Coeliac disease (CeD) is a common, underdiagnosed enteropathy with rising incidence and diagnostic delay. This literature review synthesises advances in small bowel capsule endoscopy (SBCE) and artificial intelligence (AI) for SBCE, and outlines implications for clinical practice.
METHODS: A comprehensive literature search in PubMed, Scopus, Embase, and Cochrane Library was conducted, where relevant articles published in English over the past ten years [2015-2025] were selected and analysed by two independent reviewers.
KEY CONTENT AND FINDINGS: Current evidence supports tissue transglutaminase immunoglobulin A (tTG-IgA) as the first-line test and endomysial antibody IgA (EMA-IgA) as a test to rule in disease. An adult no-biopsy pathway at ≥10 times the upper limit of normal (ULN) yields near-perfect specificity but modest sensitivity; therefore, histology remains the reference standard. Optimised biopsy protocols with ≥4 samples from the second part of the duodenum plus 1-2 samples from the bulb, which are well-oriented, increase diagnostic yield. SBCE complements oesophagogastroduodenoscopy (OGD) to map disease extent, detect complications, and guide care when biopsy is contraindicated. A positive baseline study may be prognostic. AI has progressed from per-frame villous atrophy (VA) detection (internal accuracy: 94-96%) to patient-level and severity curve methods showing high agreement with experts, enabling reproducible burden mapping. Across prospective studies and meta-analyses in mixed SBCE indications, AI assistance increases sensitivity without losing specificity and reduces review time approximately 10-12-fold. Gains are greatest for non-experts and for triage applications. Key limitations include small, single-centre datasets, inconsistent labelling, image frame analysis rather than full videos, data leakage risks, and uncertain generalisability across devices and populations. Priorities include multicentre, patient-wise external validation; harmonised International Capsule Endoscopy Research (I-CARE) lesion definitions; prevalence-aware calibration; equity-aware evaluation; and vendor-agnostic deployment.
CONCLUSIONS: AI-augmented SBCE can improve efficiency, consistency, and monitoring of CeD; however, adoption should remain human-in-the-loop and be anchored to safety protocols, including patency testing when retention risk is relevant. Equity considerations include serology-negative presentations in some populations and the need for calibrated thresholds aligned with real-world prevalence and costs.
METHODS: A comprehensive literature search in PubMed, Scopus, Embase, and Cochrane Library was conducted, where relevant articles published in English over the past ten years [2015-2025] were selected and analysed by two independent reviewers.
KEY CONTENT AND FINDINGS: Current evidence supports tissue transglutaminase immunoglobulin A (tTG-IgA) as the first-line test and endomysial antibody IgA (EMA-IgA) as a test to rule in disease. An adult no-biopsy pathway at ≥10 times the upper limit of normal (ULN) yields near-perfect specificity but modest sensitivity; therefore, histology remains the reference standard. Optimised biopsy protocols with ≥4 samples from the second part of the duodenum plus 1-2 samples from the bulb, which are well-oriented, increase diagnostic yield. SBCE complements oesophagogastroduodenoscopy (OGD) to map disease extent, detect complications, and guide care when biopsy is contraindicated. A positive baseline study may be prognostic. AI has progressed from per-frame villous atrophy (VA) detection (internal accuracy: 94-96%) to patient-level and severity curve methods showing high agreement with experts, enabling reproducible burden mapping. Across prospective studies and meta-analyses in mixed SBCE indications, AI assistance increases sensitivity without losing specificity and reduces review time approximately 10-12-fold. Gains are greatest for non-experts and for triage applications. Key limitations include small, single-centre datasets, inconsistent labelling, image frame analysis rather than full videos, data leakage risks, and uncertain generalisability across devices and populations. Priorities include multicentre, patient-wise external validation; harmonised International Capsule Endoscopy Research (I-CARE) lesion definitions; prevalence-aware calibration; equity-aware evaluation; and vendor-agnostic deployment.
CONCLUSIONS: AI-augmented SBCE can improve efficiency, consistency, and monitoring of CeD; however, adoption should remain human-in-the-loop and be anchored to safety protocols, including patency testing when retention risk is relevant. Equity considerations include serology-negative presentations in some populations and the need for calibrated thresholds aligned with real-world prevalence and costs.