Normative Modelling of Brain Volume in Multiple Sclerosis

This study demonstrates that normative modelling of brain volume using a large healthy reference cohort reveals heterogeneous, deep grey matter-centric atrophy patterns in multiple sclerosis patients that are significantly associated with disability accumulation, supporting the use of population-referenced MRI metrics for individualized phenotyping and risk stratification.

The Gist

The Big Idea: Giving Every Brain a "Personalized Ruler"

Imagine you walk into a doctor's office, and they tell you, "Your brain is shrinking." You might panic. But then they add, "Well, actually, your brain is shrinking exactly as much as a 40-year-old's brain usually does." Suddenly, the news isn't so scary.

For a long time, doctors have struggled to measure brain health in people with Multiple Sclerosis (MS). They usually compare a patient's brain scan to the average of a whole group of people. It's like trying to judge if a single apple is "rotten" by comparing it to a basket of 1,000 other apples. If the basket has some bad apples, the average looks worse, and you might miss the fact that your specific apple is actually fine—or conversely, you might miss that your apple is uniquely damaged even if the basket looks okay.

This study introduces a new way to look at brains: "Normative Modelling."

Think of it like a customized growth chart for adults. Just as pediatricians use charts to see if a child is growing too fast or too slow compared to other kids their age and gender, this study created a massive digital "growth chart" for adult brains. It accounts for age, sex, and head size to tell you: "Is this specific part of your brain smaller than it should be for a healthy person exactly like you?"

How They Did It: Building the "Perfect" Reference Library

The researchers didn't just look at MS patients. First, they built a massive library of "normal" brains.

• The Library: They gathered MRI scans from 62,444 healthy people from all over the world, ranging from 6 years old to 90 years old.
• The Map: They used this data to create a mathematical map of what a "healthy" brain volume looks like at every age for men and women.

Then, they took 362 people with MS and checked their brain scans against this map. Instead of just saying "your brain is small," they could say, "Your left thalamus (a deep brain structure) is 20% smaller than it should be for a healthy 38-year-old woman."

The Key Findings: The "Thalamus" is the Canary in the Coal Mine

When they looked at the MS patients using this new "personalized ruler," they found some fascinating patterns:

1. The "Critical Deviation" Count: They counted how many parts of the brain were "out of bounds" (too small). On average, a person with MS had nearly three times as many "out of bounds" brain regions as a healthy person.
2. The Thalamus is the Star: The most common place to find these "out of bounds" regions was the thalamus. About 25% of the patients had a thalamus that was significantly smaller than expected. You can think of the thalamus as the brain's central train station. If the station is damaged, it disrupts traffic to all other parts of the city (the brain).
3. It's Not Just One Thing: While the thalamus was the most common issue, the damage was scattered. Some patients had issues in the memory centers (hippocampus), others in the balance centers (cerebellum/putamen), and others in the visual processing areas. This shows that MS is a "patchwork" disease; no two patients have the exact same pattern of damage.

Why Does This Matter? Connecting the Dots to Symptoms

The researchers wanted to know: Does having a "smaller-than-normal" brain part actually make the patient feel worse?

• Disability (EDSS): Yes. The more "out of bounds" regions a patient had, the higher their disability score. Specifically, when the thalamus, hippocampus (memory), and putamen (movement) were smaller than expected, patients had more trouble walking and moving.
• Fatigue and Cognition: The study also found links between specific brain shrinkage and feelings of exhaustion (fatigue) and trouble with thinking tasks, though these links were a bit more complex.

The "Risk Group" Strategy: A New Way to Triage

The researchers tried a clever trick to help doctors predict the future. They looked at the three brain areas that were most often damaged in MS (the thalamus, and two visual/sensory areas).

They created a simple rule: "If your scan shows damage in any of these three key areas, you are in the 'High Risk' group."

• The Result: Patients in this "High Risk" group didn't just have worse scans; they actually got more disabled over time and had a higher chance of having a relapse (a sudden worsening of symptoms) compared to those who didn't have damage in these specific spots.

The Bottom Line: From "Group Average" to "You"

The Old Way: "On average, MS patients have smaller brains than healthy people." (This is true, but it doesn't help you specifically).

The New Way: "Your brain scan shows that your thalamus is smaller than it should be for you, and this specific pattern predicts that you might face more mobility issues in the next few years."

Why is this a big deal?
This study moves medicine from a "one-size-fits-all" approach to precision medicine. It gives doctors a tool to:

1. Spot trouble early: Even if a patient feels fine, the scan might show their brain is deviating from the norm.
2. Personalize treatment: If a patient has a specific "deviation profile," doctors might choose a treatment that targets those specific areas.
3. Understand the mystery: It helps explain why two people with the same diagnosis can have very different symptoms.

A Note of Caution:
The authors are careful to say this is a research tool right now, not something you can walk into a clinic and get tomorrow. It needs more testing to make sure it works perfectly on different types of MRI machines. But it's a giant leap forward in turning complex brain data into a simple, understandable story for each patient.

In short: They built a massive "normal brain" library to create a personalized ruler for MS patients. They found that when specific brain "stations" (like the thalamus) are smaller than they should be, it's a strong warning sign for future disability. This helps move MS care from guessing the average to understanding the individual.

Technical Summary

1. Problem Statement

Interpreting brain atrophy in Multiple Sclerosis (MS) currently relies heavily on group-level research, which lacks individualized reference standards. Conventional volumetric approaches quantify absolute differences or group averages but fail to determine whether a specific patient's brain anatomy is unusually low or within expected normative ranges for their age, sex, and head size. This gap hinders clinical translation, as clinicians lack tools to contextualize individual MRI scans against population expectations. Furthermore, while deep grey matter (e.g., thalamus) and cortical atrophy are known markers of MS progression, their heterogeneity across patients and their specific relationship to disability, cognition, and fatigue at the individual level remain poorly characterized.

2. Methodology

Data Sources

• Normative Reference Cohort: The study utilized a massive dataset of 62,444 healthy controls (HC) from seven international databases (including UK Biobank and ADNI), spanning ages 6.0 to 90.1 years (50.8% female).
• MS Cohort: Data was drawn from two Norwegian sources:
  1. The OFAMS clinical trial (omega-3 fatty acid intervention, 2004–2008) with long-term follow-up (up to 12 years).
  2. Routine hospital examinations at Oslo University Hospital (since 2012).
  • Total MS Sample: 362 patients with MS (pwMS) contributing 953 longitudinal T1-weighted MRI scans.
  • Clinical Metrics: Expanded Disability Status Scale (EDSS), Paced Auditory Serial Addition Test (PASAT) for cognition, and Fatigue Severity Scale (FSS).

Image Processing & Harmonization

• Preprocessing: T2-lesions were segmented and used to fill T1-weighted images (lesion filling) to prevent underestimation of brain volumes.
• Segmentation: Regional cortical and subcortical volumes were extracted using FreeSurfer (v6.0.0 and v7.1.1) based on the Desikan-Killiany atlas.
• Harmonization: The ComBat algorithm was applied to the training (HC) data to correct for scanner/site batch effects. Test data (MS) was not harmonized to preserve biological variance and avoid introducing artificial group differences, though sensitivity analyses confirmed robustness.

Statistical Modeling (Normative Modelling)

• Model Architecture: Multivariate fractional polynomial regression was used to model regional brain volumes as a function of Age, Sex, and Intracranial Volume (ICV).
  • Models were trained separately for males and females to capture non-linear lifespan trajectories and intercept differences.
  • Equation: $region_i = b_0 + b_1 ICV_i + b_2 age^p + u_i$
• Deviation Scoring: For each MS patient, Z-scores were calculated for every region:
  $$Z = \frac{Observed - Predicted}{RMSE}$$
  • A Critical Deviation (CD) was defined as a Z-score < –1.96 (indicating volume significantly lower than expected).
• Analysis Strategy:
  • Cross-sectional & Longitudinal: Associations between the count of CDs and clinical scores (EDSS, PASAT, FSS) were tested using linear mixed-effects models.
  • Stratification: Patients were stratified into a "risk group" if they had at least one CD in the three most frequently affected regions (bilateral thalamus, right superior parietal, pericalcarine).
  • Survival Analysis: Andersen-Gill and Prentice-Williams-Peterson models were used to assess relapse risk in the OFAMS sub-sample.

3. Key Contributions

1. Large-Scale Normative Framework: Established a robust, age- and sex-adjusted reference model for regional brain volumes using one of the largest healthy control datasets (N=62k) applied to a longitudinal MS cohort.
2. Individual-Level Phenotyping: Moved beyond group averages to quantify the heterogeneity of MS, demonstrating that while the disease affects specific regions consistently, the specific deviation profile varies significantly between individuals.
3. Clinical Stratification Rule: Developed a deviation-based risk stratification method using the overlap of critical deviations in key regions (thalamus, parietal, occipital) to predict disability progression and relapse risk.
4. Decoupling Disease from Aging: Demonstrated that normative modelling can isolate disease-specific neurodegeneration from normal aging processes by referencing individuals against lifespan trajectories.

4. Key Results

Morphometric Profile of MS

• Higher Deviation Burden: pwMS exhibited nearly 3 times the number of critical deviations compared to matched healthy controls (Incidence Rate Ratio = 2.70).
• Regional Specificity: The thalamus showed the highest overlap of critical deviations (approx. 25% of patients), followed by the right superior parietal cortex (16%) and pericalcarine cortex (14%).
• Heterogeneity: While the thalamus was the most consistently affected, deviation profiles were highly heterogeneous across patients, involving subcortical (putamen, hippocampus) and cortical (precuneus, fusiform) regions.

Clinical Associations

• Disability (EDSS):
  • A higher count of critical deviations was significantly associated with higher disability scores both cross-sectionally ($\beta=0.24$) and longitudinally ($\beta=0.07$).
  • Lower-than-reference volumes in the thalamus, hippocampus, and putamen were consistently linked to higher EDSS scores over time.
• Cognition & Fatigue:
  • Specific cortical deviations (e.g., left supramarginal gyrus) predicted PASAT scores.
  • Deviations in the right isthmus cingulate predicted fatigue (FSS).
  • However, global CD counts were not significantly associated with PASAT or FSS in longitudinal models.
• Relapse Risk: Patients with critical deviations in the stratification regions (thalamus/parietal/occipital) showed a directionally consistent increase in relapse hazard (HR $\approx$ 1.60), though this did not reach strict statistical significance ($p \approx 0.08-0.09$).

Risk Stratification

• Patients assigned to the "deviation-based risk group" (presence of CD in key regions) showed a significantly higher trajectory of disability accumulation over time ($\beta=0.13$) compared to those without such deviations.

5. Significance and Implications

• Clinical Utility: This study provides a scalable framework for individualized phenotyping. Instead of comparing a patient to a generic "normal" or a specific MS group average, clinicians can determine if a patient's brain volume is statistically abnormal for their specific age and sex.
• Precision Medicine: The identification of a "morphometric phenotype" centered on deep grey matter (thalamus) and specific cortical areas allows for better patient stratification. This could aid in early detection, monitoring of neurodegeneration independent of relapses, and personalized treatment decisions.
• Methodological Advancement: The study validates that normative modelling can disentangle disease progression from normal aging, offering a more sensitive tool for tracking neurodegeneration in MS than conventional volumetric analysis.
• Future Directions: While the authors note that further validation in independent, multi-center cohorts and integration with regulatory-approved tools are needed before clinical deployment, this work lays the foundation for using population-referenced MRI metrics as a standard of care in neurology.