One of the many AI benefits is its help for scientific research. In some cases, computer algorithms are more efficient than human capabilities, are time-saving and thus, play an important supporting role in scientific analysis.
This is one of the reasons why biologists and clinicians started using computer vision for such purpose as genetic research. In 2019, the American company FDNA, composed of specialists in bioinformatics and genetics, published an article in Nature Medicine, presenting their algorithms trained to detect genetic mutations based on faces of different people with genetic disorders. During the experiments, the algorithms, named DeepGestalt, were presented pictures of patients with various syndromes causing distortions of face features, and after training, were capable to identify one condition from another. They were also capable to detect which mutation was possible to cause the disorder.
In one of the experiments, DeepGestalt showed 91% accuracy in identifying the correct syndrome between top 10 possibilities for 502 face pictures of people with 92 different genetic disorders.
In total, 17’000 face pictures of patients with more than 200 different syndromes were analyzed, and the algorithms accuracy was approximately 65%.
Healthcare specialists already use the smartphone application developed by FDNA as aid, but indeed it cannot serve as a complete medical diagnostics solution. Today, FDNA pursue their project in cooperation with clinicians and genetic experts, helping to provide people with primary diagnosis, to have a consultation with geneticists and to provide information about disorders, including the description of symptoms.
Obviously, facial recognition technologies are used in multitude of fields in relation to biology. An open-source toolset designed for phenotyping 3D images, especially for life sciences, is available on GitHub – MeshMonk. In this toolset, 2D face images with overlapping fields of view of more than 6000 participants were converted into 3D images. The average 3D image is used as template for visualization of specific traits on any face picture. This particular toolset was chosen by groups of researchers for different purposes.
One of the experiments, described in 2019 in Nature, was focused on DNA profiling based on 3D pictures of faces, using MeshMonk technology. In this work, scientists trained the program to associate a DNA sample with faces of people of different origins, sex, age, BMI and more, starting by associated segmented features with genetic information. Matching was between 80 and 83%. Researchers expect the results of their project to be used in forensics.
MeshMonk has as well served in a study analyzing how genetics impact on face morphology, also published in Nature in December 2020.
Facial recognition has a wide application field. As technology continues to evolve, more and more use cases are possible. Yesterday’s futuristic scenarios become today’s reality, crossing the boundaries between different sciences, helping people in their everyday lives and work sometimes in very surprising ways. Indeed, new technologies have also multiple risks and are subject of ethical discussions, but it is not a reason to be afraid. New things always tend to first frighten society, especially sceptical persons, however, they move the progress and open the gates to the future.
Image credit: iuriimotov, freepik.com