Secondary Logo

Journal Logo

Evaluating the Success of Facial Feminization Surgery Through Artificial and Human Intelligence

Lu, Stephen M. MD, MDiv; Chen, Kevin MD; Fisher, Mark MD; Cheng, Roger MS; Zhang, Ben H. BA; Di Maggio, Marcelo MD; Bradley, James P. MD, FACS

Plastic and Reconstructive Surgery - Global Open: August 2019 - Volume 7 - Issue 8S-1 - p 47-48
doi: 10.1097/01.GOX.0000584468.14112.e3
Craniofacial Abstracts
Open

Northwell Health, Zucker School of Medicine at Hofstra/Northwell, Lake Success, NY

This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The work cannot be changed in any way or used commercially without permission from the journal.

PURPOSE: Male-to-female transgender patients desire to be identified as female, not only with their partners but also in public settings. Facial feminization surgery (FFS) entails highly visible changes in the facial hard and soft tissues which may affect such social first impressions. No study to date has evaluated the impact of FFS on how patients are gender-typed. To study the effectiveness of FFS, we investigated preoperative/postoperative gender typing using both (1) neural networks trained on facial features (artificial intelligence) and (2) a large public online survey (crowd sourcing).

METHODS: For both studies, standardized frontal and lateral view preoperative and postoperative images of 20 patients who completed staged FFS (combinations of frontal sinus wall setback, supraorbital recontouring, mandibular angle reduction, genioplasty, upper lip shortening, septorhinoplasty, tracheolaryngeoplasty) were used; in addition, 10 male and 10 female unoperated control patients were included. (1) For the first study, the images were analyzed by 4 public neural networks trained to identify gender. Preliminary results led us to (2) a second study, using an online crowdsourcing platform. Respondents identified the gender of the same images, randomized, with a confidence rating (1: not confident; 10: highly confident). Age and smoking status were recorded as distractants. All results were recorded and analyzed for statistical significance.

RESULTS: (1) For the “neural network study,” all 4 programs provided a gender; 2 also provided a confidence score. The networks correctly identified male and female controls 98.6% and 91.2% of the time. Preoperative FFS patients were recognized as female only 54.5% of the time, whereas postoperatively this improved to 93.7%. Confidence scores (ranging from −1: confidently masculine to 1: confidently feminine) also significantly improved from 0.27 (preoperative) to 0.87 (postoperative) (P < 0.0001), with controls of −0.91 (male) and 0.89 (female). (2) For the “crowdsourcing study,” 802 people completed the survey. Control male and female images were correctly gender-identified 99.0% and 99.4% of the time with confidence 8.9 and 9.0, respectively. Preoperative FFS patients were identified as female only 57.3% of the time; by contrast, postoperatively 94.3% were identified as female, a statistically significant improvement of 37% (P < 0.0001). The confidence rating also improved from 1.41 to 7.78 (P < 0.0001).

CONCLUSION: The success of FFS (patients more likely to be identified as female) was demonstrated by both artificial and human intelligence methods. This is the first study of its kind evaluating how machine learning and the public gender type FFS patients.

Copyright © 2019 The Authors. Published by Wolters Kluwer Health, Inc. on behalf of The American Society of Plastic Surgeons. All rights reserved.