AI in Esophageal Motility Disorders: Systematic Review of High-Resolution Manometry Studies

High-resolution esophageal manometry (HRM) is essential for diagnosing esophageal motility disorders, affecting 10%-15% of patients with dysphagia. Current interpretation via the Chicago Classification remains challenging, with interobserver variability reaching 30%-40% even among experts. Artificial intelligence (AI) has emerged as a transformative tool to automate HRM interpretation.

We aimed to evaluate current AI HRM applications and assess diagnostic accuracy, methodological approaches, clinical validation, implementation barriers, and real-world implications for gastroenterology practice.

We searched PubMed/MEDLINE, Embase, Cochrane Library, and Web of Science through November 2025, for studies using AI or machine learning to interpret esophageal HRM. Eligible studies included original research evaluating such interpretation in adults with esophageal symptoms, published in English. We excluded case reports, reviews, abstracts, and studies without outcomes. Data on AI model tasks and diagnostic outcomes were extracted. Primary outcomes included diagnostic accuracy metrics, secondary outcomes encompassing external validation performance, real-time processing capabilities, and comparison with expert interpretation. Two reviewers independently screened studies and extracted data. Study quality was appraised using QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) criteria. Given the substantial heterogeneity, we performed qualitative narrative synthesis rather than quantitative meta-analysis.

Seventeen studies encompassing 4588 patients demonstrated progressive AI evolution across 3 phases. Early studies (2013-2016, n=4) using traditional machine learning achieved 86.5%-94% accuracy for parameter extraction. Deep learning era (2018-2022, n=8) achieved breakthrough performance: 97% (95% CI 95.7%-98.3%) accuracy for integrated relaxation pressure classification, 91.32 (95% CI 87.0%-94.5%) for motility tracing, and 86% for complete Chicago Classification automation. Recent multimodal approaches (2023-2024, n=5) incorporating acoustic analysis and fuzzy logic achieved 83%-95% accuracy while reducing interpretation time from 15-20 to <2 minutes. AI systems demonstrated superior consistency with 0 intraobserver variability compared to 15%-30% among human experts. However, critical gaps emerged: 0% (0/17) of studies performed external validation, 82% (14/17) showed unclear patient selection bias, and none obtained regulatory approval. QUADAS-2 assessment identified low risk of bias in 65% (11/17) of studies for the index test domain but high concern in 100% for applicability due to lack of real-world testing.

This review demonstrates AI's transformative potential for HRM interpretation, with diagnostic accuracies reaching 97%. Real-world implications are significant, promising to enable standardized diagnostics across institutions, address the critical shortage of motility experts affecting 70% of global health care systems, and reduce health care costs by 20%-30% through an 85%-90% reduction in interpretation time and decreased repeat procedures. Beyond synthesizing existing evidence, this review brings new knowledge to the field through 3 key contributions: mapping the evolutionary trajectory from rule-based to deep learning systems, quantifying AI's superior reproducibility compared to human experts, and revealing the critical disconnect between algorithmic performance and clinical translation. Future priorities include multicenter validation trials and regulatory pathway development.

PROSPERO CRD420251154237; https://www.crd.york.ac.uk/PROSPERO/view/CRD420251154237.