One Profile to Rule Them All? A Neural Network Analysis of the Homogenizing Effect of Primary Rhinoplasty.

Abstract

Background: A goal of aesthetic rhinoplasty is to improve nasal aesthetics while preserving individuality.

Objectives: We hypothesized that primary rhinoplasty has a homogenizing effect on a patient's appearance and in a distinct manner for a given rhinoplasty surgeon.

Methods: Standardized, unedited pre- and postoperative photos, including frontal, lateral, and oblique views, of 145 Caucasian female primary rhinoplasty patients from the public websites of three independent, experienced rhinoplasty surgeons were included (Surgeon A: n=50, Surgeon B: n=50, Surgeon C: n=45). Using Amazon Rekognition, similarity between photos were calculated by comparing all preoperative photos to each other and all postoperative photos to each other. Similarity differences between pre- and postoperative views were calculated intra- and intercohort and standardized through the relative change in similarity. Paired t-tests evaluated the absolute relative changes while a simple t-test assessed the intercohort and intracohort differences.

Results: Patients were significantly more similar post surgery (17.58%, p<0.001) in the lateral (19.17%), oblique (19.02%), and frontal (14.38%) views. In the subgroups, Surgeon A similarity increased 37.95% while Surgeon B had a 24.60% increase, and Surgeon C a 17.68% increase. The intercohort similarity increase (14.15%) was significantly less than the intracohort increase (30.89%).

Conclusions: This study illustrates that primary rhinoplasty patients have increased homogeneity postoperatively, particularly within surgeon cohorts. This is the first study that quantifies the homogenizing effect of rhinoplasties using neural networks and shows the quantifiable "signature" effect of a specific surgeon on their patients' appearances as a population. This study may impact how surgeons set patients' expectations.