Artificial intelligence facilitates measuring reflux episodes and postreflux swallow-induced peristaltic wave index from impedance-pH studies in patients with reflux disease.

ABSTRACT

BACKGROUND/AIM: Reflux episodes and postreflux swallow-induced peristaltic wave (PSPW) index are useful impedance parameters that can augment the diagnosis of gastroesophageal reflux disease (GERD). However, manual analysis of pH-impedance tracings is time consuming, resulting in limited use of these novel impedance metrics. This study aims to evaluate whether a supervised learning artificial intelligence (AI) model is useful to identify reflux episodes and PSPW index. METHODS: Consecutive patients underwent 24-h impedance-pH monitoring were enrolled for analysis. Multiple AI and machine learning with a deep residual net model for image recognition were explored based on manual interpretation of reflux episodes and PSPW according to criteria from the Wingate Consensus. Intraclass correlation coefficients (ICCs) were used to measure the strength of inter-rater agreement of data between manual and AI interpretations. RESULTS: We analyzed 106 eligible patients with 7939 impedance events, of whom 38 patients with pathological acid exposure time (AET) and 68 patients with physiological AET. On the manual interpretation, patients with pathological AET had more reflux episodes and lower PSPW index than those with physiological AET. Overall accuracy of AI identification for reflux episodes and PSPW achieved 87% and 82%, respectively. Inter-rater agreements between AI and manual interpretations achieved excellent for individual numbers of reflux episodes and PSPW index (ICC = 0.965 and ICC = 0.921). CONCLUSIONS: AI has the potential to accurately and efficiently measure impedance metrics including reflux episodes and PSPW index. AI can be a reliable adjunct for measuring novel impedance metrics for GERD in the near future.