Artificial Intelligence-Assisted Ultrasound Confirmation of Laryngeal Mask Airway Placement Using Fiberoptic Assessment as the Reference Standard: A Prospective Diagnostic Accuracy Study
This prospective observational study aims to develop and evaluate artificial intelligence-based models for the assessment of laryngeal mask airway (LMA) placement in adult patients undergoing elective surgery under general anesthesia. Following LMA insertion, standardized airway ultrasound images will be obtained and fiberoptic assessment will be performed as the anatomical reference standard. Fiberoptic findings will be classified as optimal (Brimacombe grades 3-4) or suboptimal (grades 1-2). Clinical and quantitative airway ultrasound variables will also be recorded. The predictive performance of tabular, image-only, and multimodal artificial intelligence models will be evaluated for identifying optimal versus suboptimal LMA placement.
Laryngeal mask airways are widely used for supraglottic airway management during general anesthesia. Although adequate ventilation can usually be achieved after insertion, satisfactory clinical ventilation does not necessarily indicate optimal anatomical placement. Fiberoptic assessment provides direct visualization of the relationship between the LMA and laryngeal structures but is invasive and not routinely available in all clinical settings. Airway ultrasonography provides a non-invasive bedside method for evaluating LMA position. In this prospective observational study, standardized post-placement ultrasound images will be obtained in adult patients undergoing elective surgery with an LMA. Fiberoptic assessment will subsequently be performed and graded according to the Brimacombe fiberoptic scoring system. Grades 3-4 will constitute optimal placement and grades 1-2 suboptimal placement. Three prediction approaches are planned. Model A will use prespecified clinical and quantitative ultrasound variables, including age, sex, body mass index, Mallampati class, mouth opening, thyromental distance, neck circumference, dentition status, tongue thickness, skin-to-epiglottis distance, and the hyomental distance ratio. Model B will use post-placement ultrasound images alone using a transfer-learning-based image model. Model C will integrate ultrasound image features with the prespecified tabular variables in a multimodal prediction model. Model performance will be assessed using patient-level resampling procedures. The primary performance measure will be the area under the receiver operating characteristic curve (AUROC), with additional assessment of sensitivity, specificity, positive and negative predictive values, F1 score, precision-recall performance, and calibration. Images obtained from the same participant will remain within the same data partition to prevent information leakage
Inclusion Criteria: * Adults undergoing elective surgery under general anesthesia in whom LMA use is clinically planned; ASA physical status I-III; ability to provide written informed consent.
Exclusion Criteria: * Emergency surgery; pregnancy; anticipated difficult airway; major upper airway or neck anatomical abnormality or previous major neck surgery; clinically significant aspiration risk or contraindication to LMA use; inability to obtain adequate ultrasound images or fiberoptic assessment.
[{"measure":"Discrimination of Optimal Versus Suboptimal LMA Placement by the Multimodal Artificial Intelligence Model","timeFrame":"During the intraoperative assessment following LMA insertion, approximately within 15 minutes after placement"}]