diagnoses can also be made with «smart» facial analysis – breaking latest news

At first, it might look like version 4.0 of physiognomy. let’s talk aboutfacial analysis performed through machine learning: an enormous amount of data, images and videos, processed by artificial intelligence systems that manage to extract “information” from large numbers. In the medical field, face reading – and more generally of the physical characteristics of an individual – to draw useful elements in particular for a diagnosis of disease has very ancient roots. Purified of its “parascientific” elements, we find it in modern medical semiotics. Now on The Lancet Digital Health Two feasibility studies on the clinical use of facial analysis have been published.

The fields of application

In the first, Antonio Porras (University of Colorado, USA) and others developed a machine learning-based facial phenotyping model capable of analyze photographs of children and adolescents and determine the likelihood of having a genetic syndrome (such as Williams-Beuren syndrome) based on facial dysmorphology. The model presents u
average accuracy of 88% by age, sex, race or ethnicity and could prove essential for early risk assessment, particularly in low- and middle-income countries where access to genetic screening and specialist services is scarce.

In the second study, Kreshnik Hoti (University of Pristina, Kosovo) and colleagues have validated a pain rating app
based on facial expressions. Using video clips of newborns (age 2.2 – 6.9 months) subjected to vaccination, show that the results of this application are well correlated with pain intensity and could be used for early diagnosis and management of childhood pain. Both tools can be administered directly in any medical facility via a smartphone application.

Unresolved ethical questions

However, as reported by the same authors, both studies present a series of both qualitative and quantitative limits. We also have to deal with a number of unresolved ethical issues regarding these tools, more generally: those on privacy, security and consent to provide data in the first place. But also the lack of rigorous regulation of facial analysis tools and the biometric data obtained, as underlined by an editorial commenting on the two works.

“Despite the potential benefits, the field of automated facial analysis has several challenges that need to be addressed. The tools should be described transparently so that the public understands their limitations and is informed of potential negative consequences (particularly in resource-limited settings). Also, they should be identified gaps for further study, such as the need to test different demographic groups and increasing access to specialist services. «Researchers and academic editors should play an active role in evaluating the ethical implications of facial analysis studies. In the end, governments and regulatory bodies need to introduce legislation
and strict guidance to protect human rights and ensure accountability,” concludes the editorial.