A group of researchers proposed the area around the human eyes as an effective biomarker of age. After training the neural network on more than 8,000 images of human eyes, the researchers presented PhotoAgeClock, a program that predicts age from photographs of eyes with an accuracy of two years. An article describing the work was published in the Aging magazine.
External signs do not always correspond to the calendar age of a person, but they can reflect the state of his body and the influence of various external factors, both negative and positive: this can be both smoking and alcohol abuse, as well as proper nutrition and regular sports. At the same time, the so-called biological age better reflects the state of the body and can be used to assess the factors affecting the aging process outside of various diseases.
The existing methods for assessing the biological age of a person are rather complicated and are mainly based on DNA analysis. In a new work, researchers led by Eugene Bobrov from Moscow State University and the Estonian tech startup HautAI OU proposed to estimate the age of people from photographs of the area around their eyes. To do this, they trained a neural network on pairs of photographs of the upper part of the face of 8414 people and their exact age. The efficiency of the neural network was then tested on other photographs: the system most accurately predicted age from the images in which the corners of the eyes were visible. The neural network predicted the chronological age of a person with an accuracy of 2.3 years.
Despite the fact that the neural network has not learned to predict biological age, the researchers are confident that effective automatic determination of chronological age can be useful for developing such systems or improving the performance of assessment by specialists. In addition, the scientists note that changes in the skin around the corners of the eyes can also be significant parameters when assessing the aging of the body when carefully analyzed.
Big data analysis and machine learning can help predict other significant medical factors. For example, in September, American researchers found that it is possible to predict the spread of obesity using satellite images of cities: for this, it is necessary to automatically assess the local infrastructure.
Elizaveta Ivtushok
