Quantitative susceptibility mapping versus R2* for iron tracking

The choice between quantitative susceptibility mapping (QSM) and R2* relaxometry will quietly shape whether the subtle degradation of deep gray matter over eighteen or twenty-four months becomes measurable, or remains buried in methodological noise. It also explains why two research groups studying the same patient cohort can publish numerically conflicting trajectories of nigral or putaminal iron, and why the same imaging protocol reviewed by a regulatory imaging endpoint committee can pass or fail on what looks, at first glance, like the same physical quantity.

The temptation is to treat these techniques as interchangeable proxies for brain iron. They are not. They interrogate different aspects of the same tissue, derive their numbers from different mathematical operations on the same gradient-echo signal, and respond in characteristically different ways to the field strength of the scanner, the choice of analysis pipeline, and the underlying microstructure of the brain itself. Understanding those differences is what separates a biomarker that survives multi-site validation from one that quietly inflates effect sizes in a single-center study.

A biomarker that survives review is a biomarker whose physical basis, methodological assumptions, and known confounders have all been named in advance.

Physical foundations: susceptibility versus transverse relaxation

Both QSM and R2* are reconstructed from the same raw ingredient — a multi-echo gradient-echo acquisition — but they extract fundamentally different physical quantities from it.

QSM calculates the localized bulk magnetic susceptibility of tissue, expressed in parts per million (ppm). It is an intrinsic, scalar property of the material: how strongly the tissue locally magnetizes when placed in the scanner's B0. Because it is reconstructed through a series of steps that separate local dipole effects from non-local background fields, QSM aims to recover the underlying tissue parameter rather than a derived signal behavior. This distinction is subtle but consequential. Susceptibility in ppm can, in principle, be compared across sites, across field strengths, and across years — provided the reconstruction pipeline is held constant.

R2*, by contrast, measures the transverse relaxation rate — the rate at which the gradient-echo signal decays — expressed in reciprocal seconds (s⁻¹). It is not a material property in the strict sense but a signal property: how quickly the tissue's transverse magnetization vanishes in the presence of local field inhomogeneities. The signal decay that R2* captures is influenced not only by iron but also by the microscopic architecture of the tissue, the orientation of fibers relative to B0, the presence of calcium or other minerals, and the macroscopic field inhomogeneities of the scanner itself.

A non-obvious consequence follows directly from this divergence. Two voxels of tissue with identical iron concentration but different myelin content can produce similar R2* values while producing different QSM values, because myelin is diamagnetic and iron is paramagnetic, and QSM distinguishes between these two regimes where R2* aggregates their effects.

Field strength and microstructural confounders

The most consequential operational difference between these two metrics is their relationship to the main magnetic field strength of the scanner.

For the same structure, R2* generally increases as field strength rises from 1.5T to 3T to 7T. This dependence is not a defect to be corrected away — it is the physical signature of how transverse relaxation arises from local field inhomogeneities that grow with the main field. For a multi-center trial that pools data from scanners of mixed field strength, or for a longitudinal study in which hardware is upgraded mid-trial, R2* requires careful harmonization protocols, statistical adjustment, or vendor-specific calibration phantoms to remain interpretable across sites.

QSM, in its theoretical construction, reports a quantity that is independent of B0. Volume magnetic susceptibility is an intrinsic tissue property, which means that a susceptibility value of, for example, 0.15 ppm in the caudate represents the same underlying chemistry whether measured at 1.5T, 3T, or 7T. In practice, the reconstruction algorithm, the signal-to-noise ratio at lower field, and the dipole inversion approach all introduce some field-dependent variability, but the underlying physical target is, by design, a field-invariant property. A team that holds its pipeline constant therefore inherits a degree of cross-platform portability that R2* cannot match without substantial statistical scaffolding.

Another subtlety emerges when we consider the role of microstructure. R2* is sensitive to tissue microstructure broadly: myelin, fiber orientation, calcification, and partial volume effects all contribute. A voxel that contains a mixture of iron-laden neurons and myelinated axons will produce an elevated R2* that conflates both contributions. QSM, because it separates paramagnetic (iron) from diamagnetic (myelin, calcium) susceptibility contributions, can in principle be used to isolate the iron signal even in heavily myelinated white matter regions. This is why QSM has emerged as the method of choice for iron quantification in tissues where myelin co-localizes with iron, such as the cortical ribbon or subcortical white matter tracts.

Reference region sensitivity in extreme iron overload

No imaging biomarker is entirely assumption-free, and QSM is no exception. The dipole inversion that recovers local susceptibility from phase data requires a reference — an internal region whose susceptibility is treated as known. In routine clinical research this reference is often chosen to be cerebrospinal fluid, deep white matter, or the whole-brain average. The choice has been treated as a minor technical detail, until one examines what happens at the extreme upper end of iron concentration.

In a study of patients with aceruloplasminemia — a rare genetic disorder characterized by profound iron overload — median QSM susceptibility values in the deep gray matter differed substantially depending on the reference region used. When the analysis referenced whole-brain average susceptibility, the median value was 0.147 ppm. When the reference was shifted to cerebrospinal fluid, the same anatomical regions yielded a median of 0.279 ppm — nearly double. The patients had not changed; only the analytical assumption about a single reference region had.

This finding carries an important practical implication. In extreme iron overload, where susceptibility values themselves approach or exceed the typical range of reference regions, the entire reconstruction becomes exquisitely sensitive to reference selection. Algorithms and pipelines that produce concordant results in healthy controls can diverge sharply in disease cohorts at the upper bound of iron burden. For this reason, any QSM-based trial enrolling patients with advanced neurodegeneration, hereditary hemochromatosis, or aceruloplasminemia must specify the reference region in advance and report it as a methodological covariate, not a footnote. Locking that choice down in the statistical analysis plan, before unblinding, is what separates a credible iron biomarker from a reconstructable one.

R2* in these same extreme conditions faces a different problem. At very high iron concentrations, signal decay becomes so rapid that the gradient-echo signal at conventional echo times has effectively vanished before the first echo. The result is a ceiling effect: R2* measurements saturate and lose their dynamic range precisely where the disease is most severe. This is why R2* accuracy is limited at very high iron concentrations, and why the method has historically been used most confidently in mid-range iron burdens.

Clinical validation in deep gray matter and neurodegeneration

The biomarker literature has accumulated a substantial body of validation work for both methods, and the patterns of concordance and divergence are instructive.

In a cohort comparing 52 patients with hereditary hemochromatosis against 47 healthy controls at 3T, both R2* and QSM detected significantly elevated iron deposition in deep gray matter structures — caudate nucleus, putamen, pulvinar thalamus, red nucleus, and dentate nucleus. At moderate iron burdens, in tissues with relatively simple microstructure, the two methods tell a concordant story, and the choice between them rarely changes the conclusion.

The picture changes in more complex tissue and in neurodegeneration. In a 7T post-mortem study of brains affected by amyotrophic lateral sclerosis, magnetic susceptibility and R2* values in the primary motor cortex showed positive correlations with histological estimates of ferritin and with myelin proteolipid protein — but the patterns of correlation were not identical. The R2* signal tracked both iron and myelin, as expected from its biophysical basis, while QSM more selectively indexed the iron-specific contribution.

Consider the implications for trial endpoint design. A therapy that aims to slow iron accumulation in the motor cortex of ALS patients will need to distinguish between changes in iron and changes in myelin, because both can shift with disease progression. A biomarker that aggregates both will partially cancel signal — myelin loss can offset iron gain, masking therapeutic efficacy in a way that no power calculation can rescue after the fact. This is precisely the situation where the differential specificity of QSM becomes methodologically decisive, and where R2* by itself risks generating false negatives not because the disease is absent but because the metric is measuring two things at once.

The endpoint committee does not ask whether your metric correlates with iron; it asks whether your metric correlates with iron specifically, and only that signal.

Strategic selection for longitudinal trial design

The decision between QSM and R2* is rarely a binary one — most contemporary trials acquire multi-echo gradient-echo data and can reconstruct both quantities from a single acquisition — but the primary endpoint choice still carries consequences that ripple through every downstream analysis.

QSM is the preferred choice when:

• The trial spans scanners of mixed field strength, or anticipates hardware changes, because its intrinsic field-independence reduces harmonization burden.
• The tissues of interest include myelinated structures such as cortical gray matter, white matter tracts, or the thalamus, where myelin confounds R2*.
• The therapeutic mechanism specifically targets iron, and iron-specific quantification is required for regulatory or mechanistic interpretation.
• The longitudinal trajectory of interest is small — fractional changes of single-digit percent per year, the kind of subtle degradation that composite metrics struggle to resolve.

R2* remains a reasonable choice when:

• The trial is single-site or field-homogeneous, and harmonization can be tightly enforced through phantom-based calibration.
• The cohort involves moderate rather than extreme iron burdens, where the ceiling effect is not yet limiting.
• The study includes populations where QSM reconstruction has been less validated — pediatric cohorts, patients with implanted devices causing severe susceptibility artifacts, or large multi-site consortia with heterogeneous acquisition.
• A secondary supportive endpoint alongside QSM is useful for cross-validation, since concordance between the two strengthens any single-modality finding and gives the data safety monitoring board a second signal to consult.

A frequently overlooked practical dimension is the longitudinal trajectory itself. Cognitive reserve, compensatory reorganization, and the slow drift of patient health over a multi-year trial all contribute to variance that the imaging biomarker must overcome. A metric whose physical basis is more directly tied to iron — and less entangled with myelin, fiber orientation, and field strength — will generally exhibit a tighter trajectory-to-noise ratio over time, which translates directly into the sample sizes required to detect a given therapeutic effect. For an early-phase trial with limited enrollment, that arithmetic matters more than any other variable in the imaging protocol.

The choice is, in the end, an exercise in choosing which assumption to make about the biology. QSM assumes that the dipole inversion and reference region can be made stable across the trial; R2* assumes that the field strength and microstructure can be controlled or harmonized. Neither assumption is costless, and neither method is unambiguously superior in every context. But for the longitudinal neurodegenerative trial — where subtle degradation is the endpoint and iron specificity is the regulatory argument — QSM has emerged as the more defensible primary biomarker, with R2* retained as a complementary supportive measure whose agreement with QSM strengthens the mechanistic interpretation.

There is a final, more reflective point worth naming. The two methods are not competing claims to truth; they are two complementary views of the same tissue, each offering its own simplification of a biological reality that is messier than either metric admits. A translational researcher who treats them as one will write papers that are easier to interpret in the short term and harder to reproduce in the next. A translational researcher who treats them as two will plan trials more conservatively, will harmonize more rigorously, and will arrive at conclusions that other sites, other scanners, and other patients can confirm. That, ultimately, is what biomarker validation is for.