Automated Objective Assessment of Facial Rejuvenation Procedures using Machine Learning
Akash Chandawarkar, MD1, Daniel J. Gould, MD, PhD2, Navin K. Singh, MD, FACS3, W. Grant Stevens, MD, FACS4.
1Johns Hopkins University School of Medicine, Baltimore, MD, USA, 2University of Southern California, Los Angeles, CA, USA, 3Washingtonian Plastic Surgery, Chevy Chase, MD, USA, 4Marina Plastic Surgery, Marina Del Rey, CA, USA.
PURPOSE: The outcomes of facial rejuvenation procedures, such as facelift, chemical or laser resurfacing, fat grafting, and injectables are often presented with subjective assessment of age reduction or improvements in attractiveness. Previous studies have attempted objective assessment by judgment of photographs by independent members of the public. These purported objective assessments open the possibility of bias based on survey methodology and are time-intensive to apply, resulting in assessments of only small sample sets. Machine learning algorithms allow us to objectively distinguish perceived age and attractiveness without bias after training on millions of faces of the general public. We present a novel method to quickly and independently assess perceived age reduction and attractiveness after facial rejuvenation procedures using machine learning algorithms, demonstrate its feasibility, and validate results with previously published data.
METHODS: Sixty-five front-facing before and after photographs of patients who underwent facelift only (57), blepharoplasty only (0), and facelift/blepharoplasty (8) with at least 6 months' follow-up were obtained from two independent surgeons. These images were processed by Facial Features API (haystack.ai, New York, NY). Data about procedure type, gender, perceived pre- and post-operative age, and perceived pre- and post-operative attractiveness (1-10) were recorded and statistical analyses were performed using Matlab (MathWorks, Natick, MA).
RESULTS: Eighty-six percent (86%) of all 65 patients had either a reduction of perceived age or increased attractiveness postoperatively as rated by the machine learning algorithm. Patients with facelift-only had an average reduction of age of 1.87 years and an average increase in attractiveness of 0.87. Patients who had a combined facelift-blepharoplasty had an average reduction of age of 4.53 years and an average increase in attractiveness of 1.60.
CONCLUSION: We present a novel automated objective assessment of facial rejuvenation procedures using machine learning algorithms, demonstrate its feasibility, and potential pitfalls. Because the algorithm can only process front-facing two-dimensional photos, procedures that improve appearance that may be less apparent on front-facing photos (e.g. facelifting or necklifting) likely will be not as strongly detected by this method. Our study confirms this as facelift only has a lower magnitude of improvement in both age reduction and attractiveness increase than facelift combined with blepharoplasty (which is easily appreciated on frontal views). The quick and objective nature of this method enables individual surgeons to be self-critical of results, facilitates comparison of outcomes of different procedures, technologies, or surgeries, and allows us to better set patient expectations for postoperative results.
Back to 2019 Posters