Normative modeling in neuroimaging across psychiatric and neurological disorders: a systematic review and the NORMA reporting checklist
This systematic review of 122 neuroimaging studies highlights the shift toward individualized normative modeling in psychiatry and neurology, identifies critical gaps in empirical foundations and reporting transparency, and introduces the 18-item NORMA checklist to standardize model construction, validation, and clinical application.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Imagine you are a doctor trying to figure out if a child is growing up healthy. For over a century, you haven't just asked, "Is this group of kids taller than that group?" Instead, you pull out a growth chart. This chart shows the normal range of heights for every age and sex. You don't just look at the average; you look at where this specific child falls on the curve. If they are way off the chart, that's a clue something might be wrong. This is the power of "normative modeling."
For decades, brain scientists have tried to understand mental and neurological disorders (like schizophrenia or autism) by comparing the average brain of a group of patients against the average brain of healthy people. It's like comparing the average height of a basketball team to the average height of a soccer team. But this approach has a big flaw: it treats everyone in the group as if they are the same. In reality, brains are messy. Two people with the same diagnosis might have completely different brain issues, and two people with different diagnoses might share the same brain quirks. The "average" brain often hides the unique, individual story of the person sitting in front of you. This paper asks a crucial question: Can we finally move from comparing groups to understanding individuals, and are we doing it right?
The Big Picture: A Systematic Review
This paper is a massive "systematic review," which is basically a super-detailed treasure hunt. The authors, led by Shinichiro Luke Nakajima and a team of researchers, scoured the scientific literature to find every study that tried to use this "growth chart" approach for the brain. They found 122 studies that fit the bill, published between 2018 and early 2025.
Here is what they discovered about the current state of the field:
1. The Tools Are Getting Fancy
The scientists are upgrading their math tools. In the past, most studies used a method called Gaussian Process Regression (GPR). But recently, the field has shifted toward more complex and flexible methods, specifically Bayesian regression and distributional regression. Think of it like upgrading from a simple ruler to a 3D scanner that can measure not just the average height, but also how much people vary in height and shape. By 2024, these newer methods were being used in nearly 60% of the studies.
2. The "Reference Group" Problem
To build a brain growth chart, you need a huge, healthy reference group to define what "normal" looks like. The authors found that while some studies used massive datasets (up to 77,157 people!), the median study only used 642 people. That's a bit small to get a perfect map of the extremes.
More importantly, the "map" is often biased. The largest datasets, which many researchers rely on, mostly come from the UK and the US and are overwhelmingly White. The paper points out that if your reference map is drawn mostly for one type of person, the "deviation scores" (how far off the map a patient is) might be wrong for people of other backgrounds. It's like using a growth chart made for adults to measure a toddler; the numbers might look "abnormal," but that's just because the chart was built for the wrong group.
3. The "Black Box" of Validation
This is the paper's biggest concern. Just because you built a fancy model doesn't mean it works on new people. The authors found that external validation—testing the model on a completely new group of healthy people from a different place—was reported in only 8% of the studies.
Imagine a weather app that predicts rain perfectly in your city but has never been tested in a neighboring town. If you take that app to the next town, you have no idea if it will work. The paper suggests that without this step, we can't be sure if the "abnormal" brain scores we find are real or just a glitch in the model.
4. The Missing Details
The researchers also noticed that many studies didn't tell the whole story. While almost everyone reported what they were studying, fewer than half fully explained how they built their models or who exactly was in their patient groups. Did they account for medication? Did they check for other health issues? Without these details, it's hard to compare studies or trust the results.
The Solution: The NORMA Checklist
Because the field is growing so fast but reporting so inconsistently, the authors created a new tool called NORMA (NORmative Modeling Assessment). It's an 18-item checklist, like a pilot's pre-flight safety list. It forces researchers to be transparent about:
- How they built their reference group.
- How they handled differences between different scanners or hospitals.
- How they validated their model.
- How they interpreted the results for patients.
The Bottom Line
The paper concludes that while the methods for creating these individual brain maps are getting better and more sophisticated, the foundation is still shaky. The technology is ready to help doctors treat individuals, but the data and the reporting standards aren't quite there yet.
The authors suggest that until we have larger, more diverse reference groups and until researchers start rigorously testing their models on new populations (external validation), these individual "deviation scores" should be treated as powerful research tools, not yet as standalone tools for making clinical decisions. We are building the engine, but we still need to make sure the map is accurate for everyone before we hit the road.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.