85. Artificial Intelligence in the Objective Comparison of Endoscopic Browlift Techniques

Abstract

Purpose: Facial aesthetic surgeries tend to use expert opinion and patient satisfaction as a means for judging success. However, these techniques are inherently influenced by personal bias and subjectivity. This limits the objective comparison of the patient’s outcome over time or outcomes of different surgeries. This study examines the use of the artificial intelligence software: FaceReader, to objectively compare a novel and well established browlift technique. Methods: 66 patients who underwent endoscopic browlift were identified and separated into two groups. Group A (n=52 (2006-2018)) underwent standard endoscopic browlift with hairline incisions. Group B (n=14 (2016-2020) underwent a novel technique of endoscopic browlift. The novel technique involves incisions and forehead anchoring within the forehead crease lines to minimize posterior displacement of the hairline. On average 2 concomitant procedures were performed with the novel technique while on average 1.2 were completed on the pre-existing technique; the most common being blepharoplasty. Patients were excluded if their image could not be run by the software or if they had had a rhytidectomy as a concomitant procedure. This resulted in 46 analyzed patients in Group A and 12 patients in Group B. Pre- and postoperative neutral expression images were then analyzed using FaceReader. The software tracks 500 key locations on the face to accurately measure the presence and intensity of 28 action units and 7 cardinal emotions in each image. Results: Across the procedures, the only emotions the software detected to have changed significantly were perceived happiness, sadness, and anger. Group A had an increase in happiness from 1.63% to 9.45% (p=0.004) while Group B increased from 5.24% to 12.59% (p=0.18). Group A had a decrease in sadness from 14.85% to 7.98% (p=0.06), while Group B decreased from 18.86% to 10.29% (p=0.1). Most significantly anger decreased in Group A from14.76% to 5.74% (p=0.007) with Group B decreasing from 26.41% to 10.88% (p=0.07). To assess the brow position, the brow lowerer action unit results were analyzed. Both groups had significant decrease activity in the action unit: Group A decreased intensity from 17.8% to 3.8% (p=0.0005) and Group B decreased from 35% to 8.4% (p=0.011). These findings report nearly identical relative changes except for happiness which was increased 480% in Group A compared to 140% in Group B. Conclusion: This study provides the proof of concept for the use of artificial intelligence to objectively compare surgical outcomes between two browlift techniques. This study demonstrates that the novel technique of browlift is as effective at brow elevation with similar patterns of alterations in patients’ emotions. Future applications of this software should include large patient population assessment of various facial surgeries to uncover which have the greatest impact on functioning and emotional expression.