Start Over

Otolaryngologist perceptions of AI-based sinus CT interpretation.

Authors :: Massey CJ
Asokan A
Tietbohl C
Morris M
Ramakrishnan VR
Source :: American journal of otolaryngology [Am J Otolaryngol] 2023 Sep-Oct; Vol. 44 (5), pp. 103932. Date of Electronic Publication: 2023 May 21.
Publication Year :: 2023
Abstract: Background: Overcoming non-standardization, vagueness, and subjectivity in sinus CT radiology reports is an ongoing need, particularly in keeping with data-driven healthcare initiatives. Our aim was to explore otolaryngologists' perceptions of quantitative objective disease measures as enabled by AI-based analysis, and determine preferences for sinus CT interpretation.<br />Methods: A multi-methods design was used. We administered a survey to American Rhinologic Society members and conducted semi-structured interviews with a purposeful sample of otolaryngologists and rhinologists from varying backgrounds, practice settings and locations during 2020-2021. Interview topics included sinus CT reports, familiarity with AI-based analysis, and potential requisites for its future implementation. Interviews were then coded for content analysis. Differences in survey responses were calculated using Chi-squared test.<br />Results: 120 of 955 surveys were returned, and 19 otolaryngologists (8 rhinologists) were interviewed. Survey data revealed more trust in conventional radiologist reports, but that AI-based reports would be more systematic and comprehensive. Interviews expanded on these results. Interviewees believed that conventional sinus CT reports had limited utility due to inconsistent content. However, they described relying on them for reporting incidental extra-sinus findings. Reporting could be improved with standardization and more detailed anatomical analysis. Interviewees expressed interest in AI-derived analysis given potential for standardization, although they desired evidence of accuracy and reproducibility to gain trust in AI-based reports.<br />Conclusions: Sinus CT interpretation has shortcomings in its current state. Standardization and objectivity could be aided with deep learning-enabled quantitative analysis, although clinicians desire thorough validation to gain trust in the technology prior to its implementation.<br /> (Copyright © 2023 Elsevier Inc. All rights reserved.)