An Artificial Intelligence Tool for Image Simulation in Rhinoplasty

Overview

Surgeons usually create a computer simulation of the expected result during a rhinoplasty consultation. The study asked whether an artificial intelligence model can learn a surgeon's style and criteria and produce that simulation automatically. It was a cross sectional survey of resident and specialist ear, nose and throat doctors conducted in November 2019 at a rhinoplasty conference.

Sections of note

• Cross sectional survey; the study reports a level of evidence of 4.
• Ninety seven of 122 doctors agreed to participate.
• Simulation images created by a surgeon and by the artificial intelligence model were shown in random order.
• Participants rated agreement on a seven point scale, from 1 for total disagreement to 7 for total agreement.
• Median agreement with the surgeon's simulation was 6, interquartile range 5 to 7.
• Median agreement with the model's simulation was 5, interquartile range 4 to 6, P less than 0.0001.
• Evaluators totally or partially agreed with the model 68.4 percent of the time, 95 percent confidence interval 64.9 to 71.7.
• They totally or partially agreed with the surgeon 77.3 percent of the time, 95 percent confidence interval 74.2 to 80.3.

What it means for a patient

• The automated simulations were rated close to but significantly below the surgeon's own.
• The authors suggest this could give patients a realistic preview before an in person consultation.
• Doctors rated the images, not patients, and no actual surgery or outcome was involved.
• A simulation is a rendering of an intended result, not a prediction of what surgery will produce.

Why this paper matters

Simulation is central to rhinoplasty consultation and depends on the individual surgeon's time and eye. This study tests whether that judgment can be automated. Whether an automated simulation set before a consultation helps or misleads patients is untested.

Terms

• Artificial intelligence model: software trained on examples to reproduce a pattern automatically.
• Simulation: a digitally altered image showing an intended surgical result.
• Likert scale: an ordered rating scale, here with seven points.
• Interquartile range: the middle half of the values in a set.
• Cross sectional survey: a snapshot of opinions collected at one point in time.