107. Facial Feminization Surgery: Region-specific Importance Recognized by Artificial and Human Intelligence Gender Recognition Correlates with Patient Satisfaction

Abstract

PURPOSE: Facial Feminization Surgery (FFS) has emerged as an important aspect of social gender confirmation for trans women through modifications to the facial hard and soft tissues. However, there is limited information about which of the many FFS procedures are the most important. To determine this, individual FFS procedures or regional changes were compared preoperatively and postoperatively using gender typing from both artificial intelligence (neural networks trained on facial features) and public opinion (online crowdsourcing survey). METHODS: For both studies, standardized frontal and lateral view preoperative and postoperative images were used; in addition, unoperated control cis-gender males and cis-gender females were included. Part I: The outcome of 12 different individual FFS procedures (eg. osseous genioplasty vs rhinoplasty) or facial regions (eg. forehead vs chin) (n=303 patients) was comparatively analyzed using four neural networks (AI which have been trained to recognize facial images to assess gender) and crowdsourcing public opinion of gender type (n=917) to measure changes in correct gender-typing, improvement in gender typing (preop vs postop), and confidence in femininity. Part II: The nasofrontal region (frontal sinus setback/rhinoplasty) was separated into subtypes based on morphologic typing of the preoperative severity (Types 1-3) and change was compared for success in FFS using neural networking and crowdsourcing. In addition, FACE-Q surveys were used to assess patient-reported facial aesthetic outcome. RESULTS: All four neural network platforms provided a gender. For all four neural networks, cis-male and cis-female control images were gendered correctly 98% and 99% of the time respectively. Preoperative FFS patient images were recognized as female only 52% of the time. With postoperative FFS, a combination of all the procedures followed by correction of the nasofrontal region had superior outcomes (98%, 96% correct gendering) compared to other facial regions (range 68-86%). With public opinion, similar results were recorded, with a combination of all the procedures followed by correction of the nasofrontal region having superior outcomes (97%, 95% correct gendering with improved confidence levels of 8.9+1.2 and 8.1+2). For the nasofrontal region, improved outcome in correct gendering was seen with a more severe preoperative state (Type 3 brow/dorsal hump) and increased change in measured nasofrontal angle (from 90o to 135o). Finally, FACE-Q scores demonstrated a high level of patient satisfaction for facial appearance (75.1+8.1) and quality of life (82.4+8.3). There was a positive correlation between less AI and public misgendering and patient reported FACE-Q scores for appearance, quality of life, and overall satisfaction. CONCLUSION: The success of FFS (patients more likely to be correctly identified as female) was demonstrated by both artificial and human intelligence methods. For FFS feature changes, both all and nasofrontal regions should be considered the most important for less misgendering by artificial and human intelligence; these outcomes positively correlated with patient satisfaction.