Artificial intelligence-assisted tracheal intubation in humans: a prospective observational study of diagnostic accuracy.

Abstract

larynGuide™ is a novel assistive software integrated with the C-MAC^® videolaryngoscope, which provides guidance during laryngoscopy and advises on tracheal tube position. This first in-human study evaluated the accuracy and reliability of larynGuide compared with the judgment of the airway operator. This prospective, single-centre, investigator-initiated, observational study included adult patients undergoing elective surgery requiring general anaesthesia with tracheal intubation. After informed consent and standardised induction of anaesthesia, laryngoscopy and tracheal intubation were performed with a C-MAC^® videolaryngoscope with a Macintosh blade by a board-certified anaesthetist. larynGuide ran on a second screen, visible only to the study team but blinded to the airway operator. After tracheal intubation attempts, the airway operator confirmed tracheal tube placement visually and with capnography. The primary outcome was the real-time accuracy of larynGuide in identifying correct tracheal tube placement. We enrolled 132 patients, of whom 110 were analysed. Of 108 patients with correctly placed tracheal tubes, larynGuide identified 102 (sensitivity 0.94, 95%CI 0.88-0.98). In six patients, the software misclassified tracheal tube position: two false negatives (i.e. the software advised a failed tracheal intubation despite correct placement); and four patients with no feedback. Among two patients with unsuccessful tracheal intubation due to oesophageal tube placement at the first attempt, larynGuide detected one. This first in-human study has established the feasibility of AI-guided real-time tracheal intubation using larynGuide. The software showed promising sensitivity, while specificity was limited. Videolaryngoscopy image quality issues, including fogging and poor visibility, impaired the performance of the software.