AI Case Study

Researchers develop non-invasive visual biomarkers of ageing using deep neural networks

Researchers from Haut.AI collaborated with Insilico Medicine to develop a system that predicts chronological age, the PhotoAgeClock. Using a DNN-based network on 8,414 anonymized high-resolution left and right eye corner photos, the study found that the eye corner and eyelid areas are highly-accurate predictors of chronological age.

Industry

Healthcare

Healthcare Providers And Services

Project Overview

"Researchers from Haut.AI, a machine vision and artificial intelligence for skin care company, in collaboration with Insilico Medicine have developed a simple and accurate predictor of chronological age called the PhotoAgeClock. These findings resulted in a research paper titled, "PhotoAgeClock: deep learning algorithms for development of non-invasive visual biomarkers of aging," published in Aging (Albany, NY), one of the leading journals in aging research.

It was found that, specifically, the eye corner and eyelid areas had the most significant impact on age prediction. As a result of this research, it is now possible to develop anonymized systems for age prediction using only photographs of corners of the eye. This information can be used in the development of personalized medical interventions and skin care treatments for aging. Highly-accurate predictors of chronological age can also be used to evaluate the impact that various lifestyle, medical, and cosmetic interventions have on aging."

Reported Results

"The results of the study demonstrate that the most accurate, non-invasive biomarker for biological age prediction is the skin around the corners of the eye. The photographic aging clock uses recent advances in Artificial Intelligence (AI) to predict age with 2.3 years Mean Absolute Error (MAE)."

Technology

"We used Xception [24], a DNN-based model. In this neural network model all layers, except the last fully- connected (dense) layer, were initialized with pre- trained weights from the ImageNet [16] dataset. We modified the model to increase its quality in the following manner. We added skip-connections from each residual block to the dense output layer and changed the last layer to fit the regression model. We used the ADAM optimization algorithm [25] and MSE loss function for training. The best quality was reached after 150 epochs." (paper)

Function

R And D

Core Research And Development

Background

"There are many factors that influence the aging process. Unlocking these factors can lead to valuable insights into what impacts the condition and health of the human body and reveal how to minimize these impacts."

Benefits

Data

"The study used 8,414 anonymized high-resolution left and right eye corner photos to demonstrate that it is possible to achieve high quality age estimations from a small facial region."