Facial Recognition Neural Networks Confirm Success of Facial Feminization Surgery

Sir: We were glad to read that Drs. Zuo and Forrest felt that our outcome of facial feminization surgery study using neural networks was innovative and clinically relevant.1 We have seen more and more dedicated transgender care centers being established in New York, Los Angeles, Miami, Canada, and all over the world. Even in the past 18 to 24 months, our Northwell Health centers have seen a greater interest among transgender patients for facial feminization surgery consultations and a greater demand for facial feminization surgery procedures. We are now consistently performing two or three full–facial feminization surgery procedures (frontal sinus setback, supraorbital recontouring, brow lift, osseous genioplasty narrowing, mandibular angle reduction, rhinoplasty, tracheal shave, and other facial procedures) each week. As this interest grows, it is important for centers to critically look at outcomes. With their correspondence, the authors highlight important considerations about our study and suggest what future facial feminization surgery study directions should be. Our study had a relatively small sample size of 20 cis-gender controls and 20 patients who underwent facial feminization surgery. It would be interesting to conduct a multi-institutional study with a much larger sample size looking at reduction of misgendering based on before-and-after images. Since the time of our study completion, we have expanded the facial feminization surgery patient database to 65 patients that had neural network analysis and patient-reported outcome measures (FACE-Q). With this expanded sample size, we have seen similar results (preoperative misgendering, 55 percent; postoperative misgendering, 2 percent). Some facial feminization surgery patients have been reluctant to share their images with our study, but most (>75 percent) willingly consented. The authors correctly point out that many of the neural networks are not universally applicable across ethnicities. Most of the millions of facial images used to create neural networks were white/Caucasian as opposed to Asian, Latino, or African American. For our study, we chose large, publicly available neural networks that had been analyzed and critiqued; however, we understand that drawbacks remain. We fully expect facial recognition technology to continue to progressively improve. As the authors point out, standardization of images is extremely important for accurate gendering by neural networks. Some transwomen have more frequent problems of public misgendering and more frequent safety issues related to that. Other patients express that with makeup and hair; they are rarely misgendered. Despite this, these patients still have a fear that without makeup and hair they will be misgendered. For example, a transwoman opening her home door to grab her takeout food without a chance to apply makeup is at risk for a response of: “Thank you, Sir,” as opposed to, “Thank you, Miss.” Although the patient in one of the representative photographs did have differing amounts of makeup applied preoperatively and postoperatively, the majority of patients had similar amounts of makeup for preoperative and postoperative images. The authors correctly emphasized the importance of hormonal therapy before facial feminization surgery. We agree and require patients to be consistently on hormone therapy with physician documentation for at least 1 year. We actually prefer hormone treatment for more that 2 years, and if facial features are still changing with ongoing hormonal therapy, we wait. For our study, the mean time of hormone therapy was 5.5 years; no patient had less than 1 year of hormone therapy before facial feminization surgery. With hormone therapy, we often see dramatic feminine changes to soft tissue but see less dramatic changes to hard tissue. Now, with more pediatric patients, it will be interesting to see the extent of facial feature changes that occur in hormone-treated patients that are not yet skeletally mature. Finally, we appreciate the authors clarifying important aspects of the study. Although there is crossover in specialties performing various facial feminization surgery procedures, the techniques learned in plastic surgery residency and craniofacial surgery fellowships are very relevant to facial feminization surgery. We hope that facial feminization surgery becomes more prevalent and incorporated into plastic surgery and craniofacial surgery training. DISCLOSURE The authors have no financial disclosures to report in relation to the content of this communication.