(Article published in Information Technology in Spanish)
Argentine researchers achieved to diagnose
automatically if a person may or may not develop a mental disease in the
future, saving months and even years of therapy.
“One believes that human behaviour it’s
super unpredictable and complicated. The key is how simplifications are done to
a certain system” said Alejo Salles, director of Cognitive and Computational
Sciences Laboratory of the University of Buenos Aires. His research is focused
primarily on finding the answer to a single question: to which extent software
explain and predict human psyche?
At the same time, other groups of Argentine
researchers are working not only on understanding how the human mind works but also
on predicting which problems may be eventually developed. But with a
difference: Salles –doctor in physics– uses mathematical and computational
tools to analyze data from experiments with children and adults.
“There is more and more information about
social life,” Salles explained. “You can make observational experiments on huge
populations and you can make mathematical models on that. The current trend is
to make mathematical models that explain conduct; these are behaviour
patterns”.
“You design an experiment that is like a
video game. You can do it online, in a laboratory, or at schools. The game is
used to collect data.” Salles explained. “Then it's a matter of detecting
behavioural effects and trying to explain why they behave like that. Why do they
follow a certain equation.”
But, he said, despite using mathematics as
a base, it’s not an exact science. “For every deviation of rationality,
economists sought an explanation. Afterwards, other researchers said, let’s go
back to the idea of rationality but thinking in contexts of uncertainty; that
is the way we live our lives. There is a probability approach called Bayesian,
which takes a lot of this idea of thinking about probability as a degree of
belief.”
“We could think that humans are doing the
same when we draw conclusions. We are doing probability theory unconsciously
when we make decisions,” said Salles.
Diagnosis in a box
Research like Salles' study of the human
psyche generally does not focus on the individual. That's what Diego Fernández
Slezak, professor of the Department of Computing, in the Faculty of Exact and
Natural Sciences of the University of Buenos Aires, and director of the
Laboratory of Applied Artificial Intelligence. Slezak along with the student
Facundo Carrillo was one of the 24 selected groups among more than 400 that
were submitted to the latest edition of the Google Research Awards for Latin
America.
Their project’s title says it all:
“Diagnosis in a box: computational characterization of mental states”. “If you
had gone to play football and you hurt your foot,” exemplified Slezak. “Later,
you’d go to the doctor and the doctor would move your ankle. He would look at
it and say that it could be some sort of ligament strain, a sprain or a
fracture. He orders you to make an X-rays plate or resonance. The doctor
receives a report and makes a diagnosis.
The app that we developed it's like the X-rays. It would be the
laboratory of the mind that gives the doctor the possibility of seeing something
that at first sight is not detected, that completes the diagnosis with the
entire clinic.”
“It will be the doctor the one who takes
care of the patient, the one who sees the whole picture, the environment. And
if the doctors want to know whether the patient might develop a mental illness,
they have a series of indexes that help them to diagnose it. Instead of waiting
for a two years follow-up, they can detect certain patterns in one interview.
This is not an automatic diagnosis, but an augmented reality. For the doctor,
it's like a Pokemon Go of the mind," says Slezak. " Through this app,
you can indicate which direction to follow when a patient shows subtle
symptoms. It can help the specialist make a diagnosis without going through
months of procedures."
Validating theories with information technology
Slezak began his project eight years ago
with the intention of extracting abstract concepts and thought patterns while
analyzing text through computer tools. "Originally," Slezak says,
"we started by trying to validate some theories of a well-known
psychologist named Julian James of Princeton, who talked about the evolution of
consciousness and introspection throughout the history of our species, which he
found reading all the literary works of antiquity. That was the first text
analysis task we did and it really worked very well."
This result was published in Frontiers in
Neuroscience magazine and it aroused curiosity among various groups of
researchers in psychiatry. They had a large amount of data available that they
could not analyze on their own. So they offered it to the Argentine researchers
in exchange for their being able to quantify it in some way. They ended up
working with data about patients with bipolar disorder, schizophrenic disorder,
and drug users in order to quantitatively measure changes in the discourse of
these patients. And they managed to do it.
“We looked for what the psychiatrists told
us we should look for,” said Slezak. “We had very good results again, and it
really exploded after they were published. We realized that this tool for
quantifying patterns in discourse to identify mental states was powerful. We
had managed to validate a psychological and anthropological theory; we managed
to detect changes in discourse in bipolar and schizophrenic patients and drug
users."
But the line of research that would lead to
the Google prize would come when they took possession of data from psychiatric
care; patients who were declared "at risk" for their chances of
developing schizophrenia in the future. These cases were followed for two and a
half years, to evaluate their progress.
Of Out of 35 patients, six became
schizophrenic during this period. “Then,” said Slezak, “we had a unique
scenario where we had an interview when doctors classified it them as risky,
and two years later, we know exactly what the diagnosis is. So with our tools,
we aim is to help the doctor, and be able to detect subtleties that they could
not detect because they do not have the automatic quantification level that a
computer has”.
“That was the result we obtained six months
ago and that is the most important. The computer was able to predict with 100
percent accuracy, which ones were the subjects that were going to convert and
which ones were not”, Slezak said.
Automatic mental analysis
Each pathology has its peculiarities, which
only psychiatrists know. We, as computers scientists, have a round trip with
them until we reach measures and indexes that help them," says Slezak.
"And once it's developed, the analysis
is automatic. But the process of reaching an index that can make predictions
could take months of work. We need a fluid dialogue with specialists until we
can define which indexes we want the computer to detect. The app is very good
to make repeated measurements, quantify things, and find subtleties that
doctors might overlook in everyday treatment. Virtually any neuronal or
psychiatric disease that has obvious motor or speech symptoms can be detected
by a computer, "says Slezak.
What is the next step? Convert it into a
product or, as Slezak says, in "a closed package". "Today we
have three or four diseases that we detect very well in laboratory experiments.
The idea is to build a kind of web or App to use from the smartphone. Then,
when a patient enters, the psychiatrist can do a standardized interview recording
the answers with a tablet or computer. The device measures some variables, and
then returns to the doctor the index of emotional appeasement, the index of
coherence of sentences and the verbs of movement to diagnose, or not, a
disease," he concludes.
Comentarios