Diagnosing mental illness: can artificial intelligence really help doctors?


“AI and Mental Health” symposium in Caen, taking place on 29 and 30 January 2024, bringing together around twenty French and international researchers and doctors, made it possible to question the position of artificial intelligence in the daily life of the therapist.

Will AI make diagnoses instead of a specialist? Not according to computer scientist Gaël Dias of the University of Caen. As for Éric Bui, a psychiatrist specializing in post-traumatic stress, he suggests seeing mental health beyond diagnosis using JITAI (“Just In Time Adaptive Intervention”, in French “Adaptive intervention just in time”). Explanations.

630,000 combinations for post-traumatic stress diagnosis

AIs are built to mimic human cognitive functions. To better understand how AI could diagnose mental health, we first need to understand how specialists in the field do it.

Psychiatrists and psychologists rely on a common foundation: the Diagnostic and Statistical Manual (DSM), currently in its fifth edition. This link is a consensus group on mental health disorders. Therefore, therapists will look for combinations of symptoms to make a diagnosis.

As highlighted by Éric Bui, PTSD refers to 20 symptoms (such as reminiscence of the event, avoidance of places or activities reminiscent of the trauma, difficulty sleeping or even tantrums). Thus, there are more than 630,000 combinations of symptoms that can determine the diagnosis of this disorder. If we add the consideration of comorbidities, it is more than a quintillion (1018) ways to be affected by this disorder! So every patient is unique.

A model that shows the limits of artificial intelligence itself

During the symposium, Professor Gaël Dias presented an analysis of a deep learning model to aid in the diagnosis of depression. Based on the interview transcripts, the model evaluates several metrics used by therapists to diagnose depression.

The model tends to overestimate these values ​​in healthy individuals and underestimate them in positively diagnosed patients. These biases in assessment do not allow us to rely solely on AI assessment, but must be coupled with non-verbal observations of the interview.

It is important to realize that, under similar conditions, therapists make mistakes of the same nature. These AIs are therefore not disposable, but can help healthcare professionals. Dr. Éric Bui presents us with the example of a psychiatrist in a hospital who, after a few hours of service, believes in his diagnosis thanks to these AIs.

The end of the diagnostic paradigm

Based on this information, we understand that the current paradigm with diagnostics is to provide a uniform profile with a generic treatment that statistically works for the disorders observed in the patient. However, mental health care can also be seen in a different way.

With the aim of further personalizing the care pathway, psychiatrist Éric Bui introduced the symptom network principle. In 2017, psychologist Denny Borsdoom, professor at the University of Amsterdam (Netherlands), proposed the application of network theory to mental health. He was able to see that the symptoms were linked by reinforcement and that reinforcing loops were established.

Once these loops are identified, it is possible to attack the network by focusing only on the central symptoms (anxiety, withdrawal, etc.). Thus, by personalizing the treatment, it would be possible to effectively treat the disorder.

The principle of stepped care to relieve the burden on specialists

Thus, these personalizations of care seem to require a lot of work from already overburdened health professionals (for example, 9 psychologists per 100,000 population in Ainu and about 22 psychiatrists per 100,000 population in the country). However, it is possible to do this by reducing the pressure on psychiatrists and psychologists.

For this, it is possible to introduce the principle of stepped care. This principle aims to divide care into intensity: at the very bottom we find “self-care”, i.e. the daily care we take (it can be a moment of relaxation, breathing exercises or perhaps a sports session). At the very top of the list are care requiring the intervention of a therapist (drug treatment, hospitalization, etc.). The higher the needs, the more the paramedic is present.

JITAI: real-time care

For the lowest level of care that patients can theoretically perform themselves, they may be confused, not knowing what to do and when. Psychiatrist Éric Bui suggests the use of JITAI.

JITAI are applications based on data from phones and connected watches that offer intervention at the right time.

For example, if you start to feel anxious on the street, your heart rate will increase and you will begin to sweat. These physiological changes can be detected using a connected watch, and for some, even a phone. If this anxiety is detected, an alert can be sent suggesting a breathing exercise.

For Éric Buie, these JITAIs can make it possible to spread out treatment while also customizing it. It specifies that the patient must be able to suspend the technique at any time because his approval remains extremely important.

Leave a Comment