Evidence of White Matter Neuroinflammation in [ME/CFS]: A Diffusion-Based Neuroinflammation Imaging Study 2026 Yu et al

[Hutan]

3.5.1 Lower NII-RF Associated With Worse Mental Health and Increased Disability
Among all participants (including patients and HCs), significantly positive associations were observed between NII-RF and MCS or BDS across major white matter tracts (Figures 5 and 6, Table 3). Note that the regions where NII-RF significantly associated with MCS and BDS largely overlapped with the regions where ME/CFS participants exhibited significantly lower NII-RF compared to HCs.

That section of the paper reinforces that low NII-Restricted Fraction was associated with worse health. BDS is Bells Disability Scale - lower numbers mean lower function. So, a 'positive association' means higher NII-RF was associated with better function.

******
Diffusion Basis Restricted Fraction as a Putative Magnetic Resonance Imaging Marker of Neuroinflammation: Histological Evidence, Diagnostic Accuracy, and Translational Potential
That 2025 paper discusses the Restricted Fraction measure.

Here's the abstract:

Diffusion basis spectrum imaging–derived restricted fraction (DBSI-RF) isolates the low apparent diffusion coefficient water signal attributed to cellular crowding. It is therefore proposed as a putative magnetic resonance imaging (MRI) marker of neuroinflammation.

The purpose of this narrative review is to evaluate animal and human studies that compared DBSI-RF with histopathological benchmarks and clinical parameters. Across inflammatory demyelination, viral encephalitis, traumatic brain injury, and neurodegenerative disorders, DBSI-RF correlated moderately to strongly with immune cell density and distinguished inflammation from demyelinating or axonal pathology. In acute multiple sclerosis, com-bined isotropic fractions predicted lesion evolution, clinical subtypes, and deep-learning models that included DBSI-RF classified lesion subtypes with high accuracy. DBSI-RFmight also be used to track putative neuroinflammation associated with psychosocial stress, mood disorders, and anxiety disorders.

The strengths of the method include sensitivity to subclinical changes and the concurrent mapping of coexisting edema, demyelination, and axon loss. Limitations include non-specific etiology features, a demanding acquisition protocol, and limited large-scale human validation. Overall, DBSI-RF may demonstrate a promising diagnostic and prognostic accuracy, warranting standardized, multicenter, prospective trials and external validation.

Overall, DBSI-RF is hypothesized to serve as an MRI marker of neuroinflammation, since cellular infiltration and glial activation in neural tissue increase the density of cells. Thus, the fraction of the restricted diffusion compartment is also elevated [16–21].

In optic neuritis, the DBSI restricted fraction increases with the number of DAPI-counted nuclei and decreases with anti-inflammatory treatment.

It's clear from that abstract and quotes that RF is seen as correlating with immune cell density and a possible measure of neuroinflammation - more RF indicates more neuroinflammation. But, this Shan study found low RF in ME/CFS..... I don't know if the Shan et al study are reporting things differently?

The abstract of the 2025 paper also calls this measure 'promising' and warranting trials and validation in humans. Clearly, Shan and the team are using cutting edge technology, which is great, but it means it's a bit hard to know what it means.

 

[DMissa]

Good to see more Aussie research that (at least on first inspection until people who understand the details weigh in) appears to be rigorously conducted

 

[DMissa]

They are also suggesting a change in cellularity although I doubt they can pin that down to microglia, which would be the inflammatory cell. They cite NAkatomi but the signal in Nakatomi's study was not in white matter particularly. It was mostly down in midbrain and brainstem I think. And it was not replicated.

It might be that this is finally going to give us evidence of structural change in brain of a low-grade inflammatory type. However, (1) they point out the inconsistency of previous studies and this one may be no different and (2) it may be better to analyse this in terms of the raw data - water, cell and fibre changes, without premature labelling as inflammatory.

From the paper: "This study has some limitations that should be acknowledged. First, although the NII model offers biologically informed metrics, it is still an indirect measure of neuroinflammation and does not differentiate between specific inflammatory cell types or processes. Validation with other neuroinflammation-specific techniques would strengthen the interpretations"

It seems they are cognisant of this, which is great.

 

[Hutan]

Just coming back to this because it is so puzzling and I'm trying to find what I am not understanding. The Yu/Shan 2026 ME/CFS study is very clear that the Restricted Fraction is lower in people with ME/CFS.

Compared to HCs, ME/CFS patients exhibited widespread white matter abnormalities, including significantly lower NII-HR and NII-RF

The Yu/Shan study cites a 2020 study of people with obesity.

The NII model has been successfully applied to detect neuroinflammation in multiple sclerosis (Wang et al. 2011, 2015), Alzheimer's disease (Wang et al. 2019, 2024), and obesity (Samara et al. 2020).

That 2020 obesity study says in its abstract:

In both cohorts, the obese group had significantly greater DBSI-derived restricted fraction (DBSI-RF; an indicator of neuroinflammation-related cellularity)

That study claims that it is the high RF that suggests neuroinflammation in obesity.



Everything I can find suggests that it is a high Restricted Fraction that indicates increased cellularity, and that's an indicator of inflammation.
e.g. this:

Preliminary studies suggest that DBSI-derived metrics can putatively capture neuroinflammation in diseases like [Alzheimers AD] [33]. Ex vivo DBSI on human AD brain tissue, combined with immunohistochemical staining of microglia (Iba-1) and computational modeling, has shown increased RF in white matter compared to controls, aligning with microglial activation and cellular debris in AD [34]. These early findings, along with rodent models of AD [35], indicate a potential for DBSI-RF to quantify neuroinflammatory components in neurodegenerative disorders.

 

[Jonathan Edwards]

After all the discussion about glymphatics and brain water I do just wonder whether people with ME/CFS might have changes in all these measures as a result of lying flat more. It might be the opposite of normal pressure hydrocephalus where there is too much water outside the brain rather than inside. Lying flat has a siginifcant effect on hydrostatic pressures affecting water flux.

 

[ScoutB]

@Hutan I somehow ended up pondering this for a few hours this evening.
TLDR: It seems like they might be using the name "NII-RF" to mean kind of the opposite of what other papers are calling restricted fraction.

In the Kéri, S. 2025 paper you linked earlier, they say

In DBSI, water diffusion in cellular structures is modelled as a restricted isotropic diffusion fraction (DBSI-RF), typically defined by very low apparent diffusion coefficients (0.3–0.6 µm2/ms). This restricted fraction (RF) is interpreted as the volume fraction of water trapped by cells (e.g., inflammatory cells).

It sounds like roughly they are breaking the diffusion (of water while it's being pushed around by magnets) down into components going in different directions as well as an "isotropic" component that corresponds to the amount of diffusion going equally in all directions (expanding motion basically).

Then I think they interpret the spots where that coefficient of isotropic diffusion is "small" (0.3–0.6 µm2/ms) as areas where there are lots of cells locking up the water. (?)

Meanwhile, in the NII Processing section of this thread's paper, they define "NII-RF" as

NII-derived hindered fraction of restricted isotropic diffusion (D ≤ 0.3  μm2/ms)

I think their "D" is again supposed to be a coefficient telling you how much of this "isotropic" diffusion there is at a given spot. If I'm right about that, this would explain the discrepancy, as then the thread's paper is defining "NII-RF" to be <0.3 while Kéri, S. 2025 is using 0.3-0.6.

In the discussion, this thread's paper also says:

Lower [NII-RF] reflects more inflammatory cellularity, these positive associations imply that less cellular infiltration is associated with better mental health and better function.

Which I think confirms that their "NII-RF" measure is doing the opposite of what other people's RF measure is.

Very confusing! As if papers didn't already cite each other getting the results completely backwards often enough already...

 

[SNT Gatchaman]

We'll probably need to dig in to their referenced articles to try and understand the concepts behind these various NII- measures. First, here's a quick review of the basic MRI diffusion imaging (which this new technique is said to improve upon). Most of this is cribbed from Radiopaedia.

The basic idea is looking at how freely water molecules can randomly move with Brownian motion. (Remember the Hitchhiker's Guide to the Galaxy and the discovery of the Infinite Improbability Drive, with its freshly brewed cup of tea )

Water molecules in CSF have unrestricted movement (isotropic). An abscess would contain a lot of pus with dead neutrophils, so water can't move about so freely (anisotropic).

In normal brain, white matter tracts have nerve fibres running in parallel that encourage motion in the same direction. So with diffusion tensor imaging you look at the eigenvalues and eigenvectors of a 3x3 matrix (don't ask about the maths ). You can then do tractography to plot the location of normal white matter tracts or assess the health of the fibres in the tracts.

The terms used in DTI are —

Mean diffusivity

• The average magnitude of molecular displacement by diffusion
• The higher the MD the more isotropic the medium
• aka Apparent Diffusion Coefficient (ADC)
• We look for low ADC for something with very restricted diffusion, like pus or tightly packed tumour cells.

Fractional anisotropy

• Reflects the directionality of molecular displacement by diffusion and varies between 0 (isotropic diffusion, like CSF) and 1 (infinite anisotropic diffusion in a specific direction)
• Shows how directionally biased the diffusion is — high in healthy white matter where water has a clear preferred direction, but low in areas with poor structural organisation

Axial diffusivity

• "The longest eigenvector"
• In white matter tracts measures water movement parallel to the axon along its length
• Might be within the onion rings of the myelin sheath
• Intact axons encourages water to move freely alongside → higher axial diffusivity
• AD is reduced with axonal injury

Radial diffusivity

• "The average of the two shorter eigenvectors"
• Measures water movement perpendicular to the axon
• Restricted by intact myelin so should be relatively low
• Increases when myelin degrades (allowing water molecules to more easily move outward and away from the axon)
• Could decrease if further compacted myelin is laid down

 

[SNT Gatchaman]

Although DTI offers useful markers of microstructural integrity, it lacks biological specificity and cannot demonstrate the contributions of inflammation, edema, or axonal damage. Advanced diffusion models (Oestreich and O'Sullivan 2022) have been developed to address this limitation.

One such validated (Wang et al. 2014; Zhan et al. 2018) approach is the diffusion-based neuroinflammation imaging (NII) model (Wang et al. 2011, 2015, 2019, 2024; Chiang et al. 2014; Samara et al. 2020), which estimates multiple biologically informed indices to quantify inflammation-related processes, including the hindered water ratio (NII-HR, indicating extracellular tissue edema), restricted fraction (NII-RF, indicating inflammation-related cellularity) and fibre fraction (NII-FF, indicating apparent axonal density).

In addition, the model provides fibre-compartment diffusivities, axial (NII-AD), radial (NII-RD), mean (NII-MD) and fractional anisotropy (NII-FA). Unlike conventional DTI metrics, which are biassed by isotropic signals from oedema or cell infiltration, these fibre-specific diffusivities isolate the anisotropic component of diffusion, thereby improving sensitivity and interpretability for axonal injury and demyelination. These indices offer greater specificity in detecting changes related to extracellular fluid accumulation, cellular infiltration and axonal density, which are hallmarks of neuroinflammation.

Spoiler: References

Transdiagnostic In Vivo Magnetic Resonance Imaging Markers of Neuroinflammation (2022)

Diffusion basis spectrum imaging detects and distinguishes coexisting subclinical inflammation, demyelination and axonal injury in experimental autoimmune encephalomyelitis mice (2014)

Differentiation and quantification of inflammation, demyelination and axon injury or loss in multiple sclerosis (2015)

Quantification of increased cellularity during inflammatory demyelination (2011)

Quantifying white matter tract diffusion parameters in the presence of increased extra-fiber cellularity and vasogenic edema (2014)

Neuroinflammation and White Matter Alterations in Obesity Assessed by Diffusion Basis Spectrum Imaging (2020)

 

[Casterofspells]

After all the discussion about glymphatics and brain water I do just wonder whether people with ME/CFS might have changes in all these measures as a result of lying flat more. It might be the opposite of normal pressure hydrocephalus where there is too much water outside the brain rather than inside. Lying flat has a siginifcant effect on hydrostatic pressures affecting water flux.

Wouldn’t that effect be controlled for by the sedative controls used?

And afaik glymphatics are mostly active during the night when sleeping and horizontal anyways.

 

[Eleanor]

After all the discussion about glymphatics and brain water I do just wonder whether people with ME/CFS might have changes in all these measures as a result of lying flat more. It might be the opposite of normal pressure hydrocephalus where there is too much water outside the brain rather than inside. Lying flat has a siginifcant effect on hydrostatic pressures affecting water flux.

When you say flat, do you mean completely flat, head not propped up at all?

 

[forestglip]

I'm just starting with reading the abstract, but this directionality is reversed for between-group vs within-ME/CFS, right? If so, I think this would make the finding less compelling.

Compared to HCs, ME/CFS patients exhibited widespread white matter abnormalities, including [...] significantly higher NII-FF [neuroinflammation imaging fibre fraction] [...] across association, commissural and projection fibres. [...] Among ME/CFS patients, higher NII-FF was associated with lower disease severity.

 

[Jonathan Edwards]

Wouldn’t that effect be controlled for by the sedative controls used?

And afaik glymphatics are mostly active during the night when sleeping and horizontal anyways.

I don't think sedatives would affect the hydrodynamics.

The story about glymphatics is clearly very controversial - even among the international experts on brain water flux. I have in the last month corresponded with several and none of them agree with each other but all seem to agree that the original glymphatic story was wrong. The whole thing is very complicated. Movement down perivascular channels may be greater at night but the idea that this means increased clearance (of waste) looks very dubious.

The things you are li'ble....
They ain't necessarily so.

 

[forestglip]

I don't think sedatives would affect the hydrodynamics.

I think Casterofspells meant to say "sedentary", suggesting that sedentary people would be lying down just as much.

 

[Jonathan Edwards]

When you say flat, do you mean completely flat, head not propped up at all?

Not necessarily. The main thing would be not upright. When upright the venous drainage from the brain goes straight into vessels at subatmospheric pressure. Not if you lie down. One of the most basic signs students learn is the jugular venous pressure. When you lie nearly flat the blood in the jugular vein can be seen to bob up ad down. If you stand up the vein is empty.

 

[Casterofspells]

I think Casterofspells meant to say "sedentary", suggesting that sedentary people would be lying down just as much.

Whoops, yeah that’s what I meant!

 

[Jonathan Edwards]

Ah yes, sedentary. That would be different. The brain is about the same height above the heart whether you are sitting or standing. Lolling is a different matter though!

 

[Hutan]

In the Kéri, S. 2025 paper you linked earlier, they say

Yes. ScoutB explained the isotropic diffusion well:

It sounds like roughly they are breaking the diffusion (of water while it's being pushed around by magnets) down into components going in different directions as well as an "isotropic" component that corresponds to the amount of diffusion going equally in all directions (expanding motion basically).

The water that is confined inside something like a cell can't move much when pulled by a magnet, compared to water that isn't confined inside something. But, I don't think the problem of the seeming reversal in the Yu 2026 paper of what the result for the restricted fraction actually means compared to everyone else has been explained.

The 2025 Keri paper gives this chart to explain three categories of diffusion; restricted, hindered and free. The numbers along the x axis are the diffusion coefficients e.g. 0.3 um2/ms, a measure of how much the water moves when pulled by the magnet. The diffusion coefficients are not the fraction, which is shown on the y axis, labelled 'Relative signal fraction'.


View attachment 31156

NII-derived hindered fraction of restricted isotropic diffusion (D ≤ 0.3  μm2/ms)

(Also, what the Yu/Shan team have written there with the 'hindered' doesn't make a lot of sense if you want communicate clearly, because there are three categories: Restricted, hindered and Free. And it's not clear why they have used less than 0.3 for the diffusion coefficient range. but, whatever. The big problem is them saying that low RF means more inflammation.)

I think their "D" is again supposed to be a coefficient telling you how much of this "isotropic" diffusion there is at a given spot. If I'm right about that, this would explain the discrepancy, as then the thread's paper is defining "NII-RF" to be <0.3 while Kéri, S. 2025 is using 0.3-0.6.

In the discussion, this thread's paper also says:

Lower [NII-RF] reflects more inflammatory cellularity, these positive associations imply that less cellular infiltration is associated with better mental health and better function.

Which I think confirms that their "NII-RF" measure is doing the opposite of what other people's RF measure is.

I do think that it is possible that the authors have confused
1.diffusion coefficients D, where a lower coefficient means there is more 'solid stuff restricting diffusion', more confinement and therefore probably 'increased cellularity' and therefore probably 'cellular infiltration and therefore inflammation', with
2. volumes exhibiting diffusion in that range of diffusion coefficients that indicates confinement i.e. Restricted Fraction, RF.

The Keri 2025 paper suggests that everyone is saying that high volumes of RF is correlated well with histological evidence of inflammation. But these authors are saying the opposite, that the low volumes of RF in people with ME/CFS means inflammation.

I could be wrong about this, it certainly is hard to believe that an error this significant could be made. But this isn't making sense to me yet.

 

[Jonathan Edwards]

The Keri 2025 paper suggests that everyone is saying that high volumes of RF is correlated well with histological evidence of inflammation. But these authors are saying the opposite, that the low volumes of RF in people with ME/CFS means inflammation.

I could be wrong about this, it certainly is hard to believe that an error this significant could be made. But this isn't making sense to me yet.

I haven't got my head round this. I ma impressed by your attempts. I see a lot of potential confusions here. Early transudative inflammation is mostly a matter of increased free water. After that you get cells coming in and free water may go down. I wouldn't expect glial activation per se to alter water much at all. Standard inflammation models may give certain signal patterns but who knows if those signal patterns in a brain mean inflammation? Or what sort?

In general, inflammatory cell infiltration in brain tissue is a disaster. And nobody has reported it in ME/CFS (I am talking of more cells, not just busy microglia that are already there). Inflammatory infiltrates in white matter will almost certainly be accompanied by demyelination, which there isn't.

It seems to me that there are probably much more likely explanations in changes in water movement than inflammatory cell infiltration.

 

[Hutan]

There seems to have been good work done correlating observable inflammatory cell infiltration with high restricted fraction.

Where the correlations between RF and measured inflammation seem to fall down is with what is described as mild inflammation. That Keri paper mentions that. That paper is worth a read. Excerpts:

Across diverse models, from autoimmune demyelination to TBI, DBSI-RF consistently mirrors the magnitude of cellular infiltration measured in tissue (Table 1). Correlation coefficients between the DBSI-RF and quantitative histology range from 0.7 to 0.9 in high- inflammation settings, underscoring a strong concordance. In cases of milder inflammation, DBSI-RF shows only mild or no elevation, corresponding to weaker correlations with histology. This body of evidence supports the validity of the DBSI-RF as a potential biomarker of neuroinflammation, with histopathology serving as the reference standard.Life 2025, 15, 1599 8 of 17 However, further studies are needed to reveal specificity and sensitivity. Additionally, the discrepancy between MRI voxel size and the optical resolution of histopathological imaging may be a limiting factor in making direct and accurate correlations.

While many studies support DBSI-RF as a putative inflammation marker, further studies are warranted. Sample sizes in human studies have been modest, and more val-idation is needed across diverse populations and scanner platforms. Histopathological comparison data in humans are limited, based on case reports or autopsies [11]. Stan-dardizing the threshold that defines “restricted” is also questionable. Some studies used 0.3 µm2/ms for in vivo brain tissue, while others used up to 0.6 µm2/ms in ex vivo or rodent tissue [11,27,30]. Slight differences in threshold or fitting approach could affect the absolute values across studies. Thus, before DBSI-RF can be adopted widely, consensus on protocols and reference values is needed.