Artificial intelligence and early facial aging.

The project asked whether artificial intelligence can identify subclinical facial aging patterns, meaning early changes that are present before they are obvious on visual inspection, and whether those patterns can guide prophylactic non-surgical interventions.

• Funding: A Southeastern Society of Plastic and Reconstructive Surgeons Research Grant, awarded in July 2025.
• Investigators: Orr Shauly, MD, and Albert Losken, MD, at Emory University.
• Status: Completed.
• Presented: The grant report was presented at the 69th Annual Scientific Meeting of the Southeastern Society of Plastic and Reconstructive Surgeons in Marco Island, Florida, in June 2026.

Machine learning is already used to estimate age from the face. A 2025 systematic review of 27 studies found that full facial image analysis was the most common approach, followed by non-image methods that use biomarkers such as the methylome and the proteome.² Adding dynamic facial expressions improved accuracy, with a mean absolute error of 3.74 years.²

Computer vision research also studies apparent age, the age a face looks rather than its chronological age, and compares methods on public benchmarks such as LAP-2015, LAP-2016 and APPA-REAL.⁴ This project applied the same idea to a clinical question, whether patterns found early can inform prevention.

Dr. Shauly and colleagues reviewed facial scanning and imaging tools such as stereophotogrammetry, cone-beam computed tomography and optical coherence tomography, which are now used across aesthetics, oncology, orthodontics and neuropsychiatry.¹

• Most Common Uses: Aesthetic evaluation of the face, facial symmetry evaluation, malignancy detection, monitoring soft tissue changes after surgery and recording surgical measurements.¹
• Broader Uses: Predicting facepiece respirator fit and evaluating macular edema.¹
• Conclusion: Facial scanning, especially when paired with advanced software, can improve clinical practice and patient outcomes by providing accessible clinical data.¹

The review is the published groundwork for this project. It received no grant funding and is separate from the grant-funded study.¹

• Generalizability: Many machine learning studies of skin age had a high risk of bias, most often from small samples, and reviewers call for more ethnic diversity in datasets.²
• External Validation: In a 2026 review of 26 studies applying AI to patient-facing aesthetic data, only 1 reported external validation.³
• Patient Outcomes: No study in that review linked patient-generated digital data to a validated patient-reported outcome measure.³
• Prevention: Whether finding aging patterns early changes how a face ages with non-surgical care is the question this project was designed to test.