Deciphering human faces with artificial intelligence for healthcare

Abstract

Context

At a time of a rapid growth in the population of elderly individuals and at a time of decreased/pressed availability of human healthcare-resources, automated face analysis has the potential to offer efficient and cost-effective methods for monitoring of a number of pathologies.

Objectives

The author revisits works in automated face analysis, which have focused on designing computer vision algorithms deducing the health state of individuals. Current limitations and benefits are discussed, placing emphasis on the potential that such technology can bring.

Methods

Computer vision algorithms, most recently based on deep neural networks have been trained with facial images or videos, jointly with health state annotations from clinical experts, in order to learn such algorithms to deduce facets of health states. Examples of such notable algorithms include approaches detecting stress, depression, apathy, pain, neurological disorder, as well as classification of expressions and phenotypes of genetic disorders.

Conclusions

Such algorithms are evolving rapidly, providing increasingly reliable accuracy and can support clinicians by providing objective measures.