Enhancing the diagnostic yield of esophageal manometry using distension-contraction plots of peristalsis and artificial intelligence.

Our prior study reveals that the distension-contraction profiles using high-resolution manometry impedance recordings can distinguish patients with dysphagia symptom but normal esophageal function testing ("functional dysphagia") from control subjects. The aim of this study was to determine the diagnostic value of the recording protocol used in our prior studies (10-mL swallows with subjects in the Trendelenburg position) against the standard clinical protocol (5-mL swallows with subjects in the supine position). We used advanced machine learning techniques and robust metrics for classification purposes. Studies were performed on 30 healthy subjects and 30 patients with functional dysphagia. A custom-built software was used to extract the relevant distension-contraction features of esophageal peristalsis. Ensemble methods, i.e., gradient boost, support vector machines (SVMs), and logit boost, were used as the primary machine learning algorithms. Although the individual contraction features were marginally different between the two groups, the distension features of peristalsis were significantly different. The receiver operating characteristic (ROC) curve values for the standard recording protocol and the distension features ranged from 0.74 to 0.82; they were significantly better for the protocol used in our prior studies, ranging from 0.81 to 0.91. The ROC curve values using three machine learning algorithms were far superior for the distension than the contraction features of esophageal peristalsis, revealing a value of 0.95 for the SVM algorithm. Current patient classification for esophageal motility disorders, based on the contraction phase of peristalsis, ignores a large number of patients who have an abnormality in the distension phase of peristalsis. Distension-contraction plots should be the standard for assessing esophageal peristalsis in clinical practice.NEW & NOTEWORTHY Our findings underscore the superiority of distension features over contraction metrics in diagnosing esophageal dysfunctions. By leveraging state-of-the-art machine learning techniques, our study highlights the diagnostic potential of distension-contraction plots of peristalsis. Implementation of these plots could significantly enhance the accuracy of identifying patients with esophageal motor disorders, advocating for their adoption as the standard in clinical practice.