JMIR Publications

ABSTRACT

Background:

High-resolution esophageal manometry (HRM) is essential for diagnosing esophageal motility disorders, yet interpretation (e.g., via the Chicago Classification) remains challenging with significant inter-observer variability. Artificial intelligence (AI) has emerged as a potential tool to automate HRM interpretation and improve diagnostic consistency.

Objective:

To systematically evaluates current AI applications in HRM, assessing diagnostic accuracy, methodological approaches, clinical validation, and implementation barriers.

Methods:

We searched PubMed/MEDLINE, Embase, and Cochrane Library through Sep 2025, for studies using AI or machine learning to interpret esophageal HRM. Data on AI model tasks, accuracy, and diagnostic outcomes were extracted. Primary outcomes included diagnostic accuracy metrics, with 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 criteria.

Results:

Seventeen studies met inclusion criteria with at least 4,588 documented patients. Early studies (2013-2016) focused on automated parameter extraction using traditional machine learning, while studies from 2018 onwards increasingly adopted deep learning architectures achieving 83-97% accuracy. Recent studies (2023-2025) have explored multi-modal approaches including acoustic analysis and fuzzy logic frameworks. Notable achievements included 97% accuracy for IRP classification, 91.32% for motility tracing, and 86% for Chicago Classification automation. External validation was not performed in any of the included studies, with all 17 studies relying solely on internal validation methods. QUADAS-2 assessment identified low risk of bias in approximately 65% of studies, with concerns primarily related to patient selection and lack of external validation.

Conclusions:

AI applications in HRM demonstrate high diagnostic accuracy and potential for clinical implementation, particularly for automated Chicago Classification and pattern recognition. However, significant barriers remain including limited external validation, lack of regulatory approval, and insufficient real-world outcome data. Future priorities should focus on multi-center validation studies, development of explainable AI models, and demonstration of clinical utility through prospective trials. Clinical Trial: The protocol of this systematic review was registered at PROSPERO (1154237) before initiation of this study.