Comparisons of the reliability of airway measurements on cone beam computed tomography scans among human raters and a convolutional neural network.

OBJECTIVE: To evaluate the performance of commercially available AI tools in airway evaluation in clinical conditions. STUDY DESIGN: 100 anonymized cone beam computed tomography datasets obtained from the records of the University of Iowa were oriented and analyzed by 2 calibrated oral and maxillofacial radiology residents using InVivo software (InVivo6). Measurements made were total airway volume and minimum cross-sectional surface area of the airway. These measurements were then compared to those produced by uploading the same datasets to Diagnocat AI software (Diagnocat). RESULTS: The results indicate that all comparisons showed good to excellent reliability (ICC > 0.75), suggesting high agreement between raters. Specifically: The intraclass correlation coefficients of 0.954 (CI: 0.757-0.983) and .944 (CI: 0.732-0.978) between the human and Diagnocat measurements indicate excellent reliability. CONCLUSION: Diagnocat (Diagnocat) artificial intelligence software is capable of analyzing patients' oropharyngeal airway volume and minimum cross-sectional area in cone beam computed tomography datasets with minimal to moderate motion artifact at a level comparable to qualified human practitioners.