AI and Me: A Personal Journey to Revolutionising Mental Health
- Paris Lalousis
- 2 hours ago
- 5 min read
It was November 2004 when I received my comorbid (that is, two disorders together) psychiatric diagnosis: Panic Disorder with Agoraphobia and Major Depressive Disorder. At age 14, this had been a huge relief and also the first time I encountered what the International Classification of Diseases (ICD) and the Diagnostic and Statistical Manual of Mental Disorders (DSM) were. For the first time, I had a valid explanation for my feelings of impending doom, my breathlessness and dizziness, and my rapid heart rate any time I mustered the motivation to venture more than 10 meters away from my home. I finally had an explanation for why all I wanted to do was stay in bed all day and why I had problems sleeping during the night.
Unfortunately, this diagnosis had come 8 years too late. My memories of my symptoms as a child are less crystallised, but I remember that going to friends’ birthday parties was always a source of anxiety. The loud music, the crowded rooms, and the intense lights were all too much for me. At age 6, I wasn’t able to articulate my symptoms of panic attacks very well, so I described them as dizziness. I still tried to attend parties every year, but they would always result in the same outcome. I would call my parents telling them I felt dizzy, and they would come and take me home after 30 minutes or so. The same “dizziness” would also occur in the classroom a few times, as well as during long trips. When my parents took me to the doctor, the first course of action was a blood test and a physical examination. Nothing wrong was found so the doctor suggested that I might have a lack of vitamins contributing to my “dizzy spells”, and therefore I should take some supplements.
And so, I did. The “dizzy spells” would come and go, and for the next 8 years I would just accept them. It was not until age 14, when my symptoms became severe enough that I had to quit school for a year, that I was able to describe my symptoms more accurately: and a visit to a psychiatrist, rather than a physical health clinician, seemed like the right place to turn.
And so, in November 2004, when I received my diagnosis, my symptoms were severe enough to warrant medication – benzodiazepines for my panic attacks for a couple of months and antidepressants for my depression, which could also act as a longer-term solution for my panic attacks, alongside CBT. It took a few attempts to find the right dose of medication. After years of highs and lows, remissions and relapses, I was able to get back into education and regain good social functioning.
These experiences have formed a lot of my motivation for the research I carry out today. When I think back to those years, I have a few questions: What if I could have been diagnosed at age 6 rather than at age 14? Would I have had a less strenuous path to recovery? What if, at age 14, I could have had a test that determined the right dose of medications, at the right time? Would I have been able to get back to being myself faster? Why did I have those experiences? I hadn't experienced trauma; I had a stable and loving upbringing, so why did I have a comorbid psychiatric diagnosis?
From patient to researcher, I now try to answer these questions myself. In my work, I try to use Artificial Intelligence (AI) approaches to improve pathways for personalised medicine in mental health. My area of work is broadly structured around three foundational themes: Can we predict the onset of symptoms/a disorder before they occur? Can we better understand the mechanisms of disorders across the diagnostic spectrum? Can we provide patients with personalised options that increase patient choice?
A good example of the work that I do was published in Biological Psychiatry in 2022. I have always been fascinated by the comorbidity in the same individuals of both depression and psychosis, because much of modern psychiatry is based on the (mis)understanding of how these disorders differ, starting in the late 19th century. Emil Kraepelin, a German psychiatrist working under the leadership of Alois Alzheimer, dichotomised psychoses into manic-depressive illness and dementia praecox (the precursor of schizophrenia). Much of modern psychiatry is based on that original dichotomy. However, such clear-cut dichotomies rarely exist in mental illness.
Comorbidity rates in psychiatry are very high and follow a rule of 50%: half of people who meet diagnostic criteria for one mental health disorder also meet diagnostic criteria for a separate disorder at the same time; half of people who meet diagnostic criteria for two disorders at the same time also meet diagnostic criteria for a third disorder; and so on. Moreover, there is a lack of accepted biological or genetic markers for diagnostic categories. It is therefore very important to understand whether the current diagnoses we have reflect clinical reality and whether they have a strong biological basis.
In my 2022 Biological Psychiatry article, I used AI to try to identify whether the diagnostic categories of depression and psychosis are rooted in biology and whether a biology-first approach could be better. I used brain scans detailing the brain structure of patients with depression and psychosis and fed those into an AI algorithm. I tasked the algorithm with finding groups of similarity based on brain structure without telling the algorithm which patients had psychosis, and which had depression. If the algorithm placed most of the people with psychosis in one group and most of the people with depression in another group, then that would show us that our current frameworks are biologically based. If the algorithm identified groups consisting of a mix of patients (so-called transdiagnostic groups), then our current frameworks would be failing to capture meaningful biological pathways.
Our results showed exactly that. The algorithm identified two transdiagnostic groups. Moreover, these groups showed specific symptoms that are not usually associated with patients who belong to either group. These findings showcase a simple truth that exists in current clinical practice: while two people might have the same diagnosis, their neurobiology might be very different – yet they will be treated with the same medication; and while two people might have a different diagnosis, their neurobiology might be very similar – yet they will be treated with different medications.
Taking things a step further, I wanted to see whether predicting symptoms with AI in these new, more biologically grounded groups could offer better insights. Since we had data from these patients nine months after admission, I tried to predict their symptoms using only their data at presentation. I found that I was able to predict their 9-month symptoms more accurately in this new biologically based separation compared to the traditional depression-psychosis separation. Astonishingly, we could do that using only data from a blood test and a few questions.
Mental health research still has some way to go before we are able to get patients in the clinic, ask them a few questions and/or run a few tests, feed those into an AI algorithm, and provide accurate diagnoses, prognoses, and medication doses. However, what we are seeing in this line of work is promising and makes me hopeful.
My hope is that, in a few years, a 6-year-old Paris presenting with “dizziness” symptoms can have a blood test and get an accurate diagnosis and course of action. Or that a 14-year-old Paris receives a biologically based diagnosis rather than a comorbid diagnosis, ensuring he can receive the right treatment at the right time.
