Work

A Combined Approach to Predict Tobacco-Induced Facial Aging Using Dermatologist Knowledge Elicitation and Generative Models

Year: 2025

Type: article

Abstract: This research presents a novel, scientifically validated facial aging simulator. It integrates two published methodologies for personalized aging predictions, particularly concerning the impact of tobacco. This contrasts with widespread simulations on social network and gaming applications, often lacking scientific rigor and transparent methodology. This simulator combines elicited knowledge from 28 expert dermatologists with an AI-powered image generation system. Based on this knowledge, the model predicts 15-year facial aging signs according to various intrinsic and extrinsic factors, providing personalized probabilities of reaching specific aging stages (e.g., wrinkles, pigmentation). This work couples these 15-year predictions with a machine learning model (AMGAN) that generates personalized facial aging simulation images. The AMGAN was initially trained on a dataset of 600 individuals with sign scores averaged from 15 expert graders. We present predicted independent probabilities of reaching grades for different facial aging signs as a function of cumulative tobacco consumption. A simulation was performed using a 43-year-old subject’s facial image to demonstrate the tool’s capabilities in illustrating the impact of tobacco consumption. Confounding variables such as sun exposure, sunscreen use, and body mass index (BMI) were controlled, while cumulative smoking (less than 10 or greater than 20 pack-years) was varied. This tool effectively visualizes the known effects of smoking on aging and provides personalized quantification of its impact on specific facial wrinkles. Recognizing that individuals prioritize different facial signs, the simulator allows users to focus on areas of personal concern. By focusing on preserving skin longevity and preventing premature aging, the simulator, which can quantify the impact of smoking on individual aging trajectories, effectively motivates positive lifestyle changes. This personalized approach offers a promising preventative messaging strategy, particularly for audiences resistant to traditional methods, and strengthens the scientific rationale regarding the impact of tobacco on specific aging signs. This work introduces a novel, scientifically robust facial aging simulator, addressing the lack of scientific and dermatological validation in most existing aging simulators, especially those used in social networks and gaming applications. Integrating two published research and leveraging dermatologist knowledge, the simulator generates personalized aging predictions (visualized as images and probabilities). This personalized, quantified, and visual approach, focused on the impact of tobacco, empowers informed decisions through probability comparisons of individual facial signs, motivating behavior change. Consequently, by focusing on preserving skin longevity and preventing premature aging, it offers a novel preventative messaging tool, particularly effective for those unresponsive to traditional campaigns, by promoting healthier behaviors. The synergy between expert dermatological knowledge and AI-generated imagery reinforces the scientific basis for the impact of smoking on facial aging.

Source: Dermatology and Therapy

DOI: 10.1007/s13555-025-01570-4