Detection of Baseline Emotion in Brow Lift Patients Using Artificial Intelligence
The aesthetic significance of the periorbital region of the human face has been appreciated since ancient times. The widespread popularity of techniques such as glabellar chemodenervation, brow lift, and blepharoplasty speaks directly to the importance that patients place on this region of the face. In the plastic surgery literature, most of what is considered an aesthetically pleasing outcome is based on instinctive analysis of the aesthetic surgeon, rather than on patient assessments, public opinions, or other objective means of evaluation.
We tested a commercially available artificial intelligence system trained and validated using the Amsterdam Dynamic Facial Expression Set, which contains highly standardized images of different emotional expressions. Using machine learning, this artificial intelligence system can analyze image data and provide an objective measure of facial expression, generating a relative breakdown of each of the 7 basic emotions as well as the action units associated with these emotions. We employed this system to measure the impact of the brow lift and associated rejuvenation procedures on the appearance of emotion while the patient is in repose.
Methods/Technique
:We retrospectively identified all patients who underwent bilateral brow lift for visual field obstruction between 2006 and 2019. We excluded those patients without postoperative photos, ones with post operative photos without resolution of bruising, those with less than 6 weeks of post op photos, those with concomitant diagnoses such as facial paralysis, those with any Botox injection within the last 6 months prior to the pre or post operative photos, and patients who underwent complex head and face reconstructive procedures in addition to brow lift.
Images were analyzed using a commercially available facial expression recognition software package (FaceReader™, Noldus Information Technology BV, Wageningen, Netherlands). The data generated from the software reflected the proportion of each emotion expressed for any given facial movement and the action units associated.
Results/Complications
:A total of 52 cases were identified after exclusion. We excluded 1 patient with unilateral brow lift, 2 patients with facial paralysis, 1 patient with multiple facial reconstruction procedures, 7 with less than 6 weeks post-operative photos or significant bruising in the last available postoperative photo, 3 patients with Botox injection, and 2 with no standardized postoperative photos. The endoscopic brow lift approach was used in 50 patients and the direct approach was used in 5 patients. 50 patients also had concurrent upper eyelid blepharoplasties, 9 had upper eyelid ptosis repair, 4 had canthopexies, 5 had facelifts, 5 had fat grafting to the face, and 14 had lower eyelid blepharoplasties. Average age at presentation was 61.6 years old (range: 36.5-86 years). Post-operative photos were obtained on average 53 weeks after the brow lift (range: 42 days to 11 years).
Emotion detected:
Preoperatively, the angry emotion was detected on average of 13.06%, the happy emotion averaged 1.68%, sad emotion averaged 13.06%, the scared emotion averaged 3.53%, and the surprised emotion averaged 0.97% among all the patients.
Postoperatively, the angry emotion decreased to 5.42% when averaging among all the patients (p=0.009). The happy emotion increased to 9.35% (p=0.0013), while the sad emotion decreased to 5.42%. The scared emotion remained relatively the same at 3.4%, and the surprised emotion increased to 2.01%, however, these were not statistically significant.
The action unit (AU) changes correspond to what can be expected after brow lift surgery, with less brow lowering AUs and more brow raiser AUs.
Conclusion
:This study proposes a paradigm shift in the clinical evaluation of brow lift and other facial aesthetic surgery, implementing an existing facial emotion recognition system to quantify changes in expression associated with facial surgery. The use of software such as these opens the way to simple, real time, and objective evaluations of outcomes that are relevant for both surgeons and patients.