About
I’m a postdoctoral researcher in precision psychiatry: the effort to tailor psychiatric diagnosis and treatment to the individual patient, rather than to the average of many. I hold positions at Amsterdam UMC, Department of Psychiatry, and at Leiden University, Institute of Education & Child Studies.
The broad question I keep returning to is a simple one that turns out to be very hard.
Can we measure something in an individual’s brain that helps explain, predict, and ultimately improve their diagnosis and treatment?
Answering it takes brain scans from thousands of people across many countries, far more than any single study or hospital can collect, combined with machine learning to find patterns that hold at the level of the individual patient.
What I work on now
I work across obsessive-compulsive, anxiety and mood disorders.
At Amsterdam UMC I work with Prof. dr. Odile van den Heuvel on brain development and obsessive-compulsive disorder. We use Generation R, a Dutch study that has followed thousands of children from before birth with repeated MRI scans, to ask how differences in brain development relate to the emergence and persistence of symptoms through childhood and adolescence.
At Leiden University I work with Dr. Moji Aghajani on brain development and clinical anxiety in young people, largely within the ENIGMA-ANXIETY consortium.
Across both positions, the aim is to keep the individual in the picture, and to model the differences that usually get averaged away. The research page goes into what that involves in more detail.
From biomarkers to brain development
I completed my PhD in 2024 at Amsterdam UMC, Department of Psychiatry, supervised by Prof. dr. Guido van Wingen and Prof. dr. Damiaan Denys.
Psychiatric diagnosis rests on the subjective assessment of symptoms. Patients sharing a diagnosis can look very different from one another, symptoms overlap heavily across disorders, and treatment guidelines largely follow a one-size-fits-all logic, which for some patients means delay, or a treatment that was never going to work. The thesis asked whether machine learning applied to neuroimaging data could produce generalisable biomarkers to help with this.
The answer was more interesting than a straightforward yes or no. Findings that looked convincing in one dataset often weakened when tested on data from new hospitals, and much of what made prediction hard was not noise but the sheer variety among patients themselves. That experience has shaped how I work since: large samples, rigorous validation, reproducibility, and a lasting interest in why patients differ so much from one another in the first place.
Consortium work
Much of what I do happens through international research consortia, in which dozens of hospitals and institutes pool their brain scans to answer questions none of them could answer alone. I’ve led and coordinated analyses within ENIGMA, across both the ENIGMA-OCD and ENIGMA-ANXIETY working groups, and within GEMRIC, a global collaboration studying electroconvulsive therapy, together spanning more than seventy sites worldwide.
For ENIGMA-OCD I developed the consortium’s first shared mega-analytic framework for functional MRI, the kind of scan that measures brain activity rather than structure, across all of its sites at once. That framework has since been adopted by other ENIGMA working groups studying different disorders.
This kind of work often looks unglamorous, but it is what allows the field to move forward: agreeing on shared protocols, reconciling data collected in different ways at different hospitals, and building analyses that hold up across populations rather than fitting the quirks of one scanner in one place. The studies I have led within these consortia are described on the research page.
My collaboration with Prof. Paul Thompson, director of ENIGMA, has been supported by three personal grants from the Royal Netherlands Academy of Arts and Sciences (KNAW): the Van der Gaag, Van Leersum and Ter Meulen grants. Together they have funded repeated research visits to his lab at USC. The KNAW published a short interview about the Van der Gaag project (in Dutch), covering the use of AI and normative modelling to study brain development in young people with anxiety and mood disorders.
Approach
I care most about validation: checking that a finding still holds outside the data it came from. It is surprisingly easy to build a model that predicts well for the patients it was developed on and then fails for patients at a different hospital, scanned on a different machine. A model like that has not really learned anything about the disorder.
So the test I hold my own work to is whether it survives contact with new, unseen patients who look like the ones a clinician actually sees: people already on medication, people with more than one diagnosis, people who are mildly unwell and people who are severely unwell. Carefully selected study samples tend to leave most of that out.
I also share my analysis pipelines and preprocessing tools openly on GitHub, for much the same reason: methods that can’t be inspected or reused are hard to trust.
A full list of my publications is generated automatically from my ORCID record. Interviews and talks are on the talks and media page, and my CV is available to download.