2024-03-15 16:00:39
Worldwide, 264 million people suffer from depression, a disease that in countries like Spain, where common mental disorders cost 2.2% of GDP, is the first cause of disability attributed to a single disease. To combat it, early detection and subsequent monitoring is key, but its diagnosis faces two major problems today: the absence of objective measures and the current overload of public health.
In order to try to mitigate them, and taking advantage of the latest advances in artificial intelligence (AI) in terms of natural language processing and computer graphics, the LabLENI of the Polytechnic University of Valencia (UPV), which has been working for years on the generation of virtual humans, is carrying out two joint projects, REMDE and DEBIO -financed respectively by the UPV and the Ministry of Social Rights and Agenda 2030 of the Spanish government- that aim to develop a virtual reality application that allows the early detection of depressive symptoms through gaze and voice biomarkers during interactions with virtual humans (DEBIO) and delve deeper, along the same lines, into neurophysiological biomarkers (REMDE).
Virtual humans as a means of stimulation to recreate realistic conversations
“The general origin of the project,” explains Mariano Alcañiz, director of LabLENI (Human Tech UPV) and principal researcher at DEBIO, “is, through a new technology such as virtual humans, to improve the tools that we make available to staff. that
works in mental health, so that they can carry out a more accurate and faster diagnosis, in general, of mental health disorders, and specifically, of depression disorders.” This is one of the objectives of the recently approved Mental Health and Addictions plan (2024-2027) of the Generalitat Valenciana.
Specifically, as indicated by Javier Marín, professor of the Department of Statistics and Applied Operational Research and Quality, member of LabLENI and principal investigator of REMDE, “we are working on the creation of virtual humans that stimulate subjects with conversations in realistic situations, of “so that, through these conversations, certain behavioral or neurophysiological biomarkers may appear, which allow us to distinguish people with depressive symptoms from those who do not have them.”
“Once the behaviors and responses of the subjects during the conversations have been measured,” continues Marín, “the second part of the project consists of modeling the patterns through machine learning and thus being able to recognize these symptoms automatically.”
The UPV has demonstrated the usability of virtual humans on the path towards automatic recognition of depressive symptoms. (Image: UPV)
Validated prototype
To validate the virtual human prototype, the research team – made up of José Llanes, Lucía Gómez, Alberto Altozano, Eleonora Minissi, Francesca Mura, Jose Roda and Carmen Calero together with Mariano Alcañiz and Javier Marín – has carried out an experiment with 100 people, of whom half presented depressive symptoms.
“The prototype has passed a first technical validation, testing its usability,” says Marín. “It has been confirmed that the conversations have a high degree of realism and naturalness, and are capable of modulating the emotions of the subjects,” adds the researcher belonging to the University Research Institute of Human-Centered Technologies, of the UPV. The results have been published in the academic journal Expert System with Applications.
Preliminary biomarker results
Now, the UPV team is working on modeling the biomarkers. In this sense, the preliminary results published at the International Conference on Affective Computing and Intelligent Interaction, held at the Massachusetts Institute of Technology (MIT) in the United States, says Marín, “indicate that subjects with depressive symptoms presented eye patterns with blinks shorter and longer saccadic movements – rapid eye movements between two fixation points – as well as forward and backward movements, possibly related to greater stress, in addition to avoiding eye contact.
“At the speech level,” he continues, “participants with depressive symptoms used more words associated with denial and exclusion, and referred more to negative emotions. Likewise, at a neurophysiological level, they showed less sympathetic activity – noted in relation to heart rate variability – and electrodermal activity.”
Finally, it should be noted that, among the next steps of the research, in the coming months a validation of the prototype with a clinical sample will be carried out in collaboration with Yolanda Cañada and Luis Miguel Rojo, from the Hospital Universitario y Politécnico la Fe, and Jon Iñaki Etxeandia, from the Valencia University Clinical Hospital.
“In short,” concludes Marín, “we are faced with reliable, objective and standardizable bioinformatics tools, which can be of great help for the early detection of affective disorders and their subsequent monitoring to reduce relapses.” (Source: UPV)
#Virtual #humans #automatic #recognition #depressive #symptoms