MRI Diffusion Tensor Imaging: Acquisition and QC

Diffusion tensor imaging (DTI) turns MRI from a picture into a measurement—fitting a three-dimensional model of water diffusion in every voxel to quantify both how much and in which direction water moves through tissue. From that tensor come the maps clinicians and researchers rely on: fractional anisotropy (FA), mean diffusivity (MD), and the fiber orientations that drive white-matter tractography.

Because DTI produces numbers that inform diagnosis, surgical planning, and research conclusions, the acquisition and the quality-control program behind those numbers are as important as the sequence itself. A FA value is only meaningful if the signal-to-noise ratio, b-value, gradient directions, and distortion corrections are all under control. This guide works through the tensor physics, the acquisition requirements, and a defensible DTI QC program. DRPS supports MRI programs as part of its MRI physics testing and quantitative-imaging services across Florida, Maryland, Virginia, Washington DC, California, Nevada, Pennsylvania, New York, New Jersey, and Delaware.

Introduction

All diffusion MRI exploits a simple idea: water molecules in tissue are in constant random (Brownian) motion, and during their microscopic wanderings they probe structure far below the image resolution. ¹ By applying a pair of strong magnetic-field gradients, the scanner makes the signal sensitive to this motion—spins that move between the two gradients lose phase coherence and signal. The more water diffuses along the gradient direction, the more the signal drops.

Diffusion-weighted imaging (DWI) measures this signal loss and reports a single apparent diffusion coefficient (ADC) per voxel, implicitly assuming diffusion is the same in every direction. But in organized tissue—most importantly brain white matter—diffusion is anisotropic: water moves more freely along axon bundles than across them. ^{2, 3} A single ADC cannot capture that directionality. DTI solves the problem by modeling diffusion as a tensor, a mathematical object that describes diffusion magnitude in every direction at once. ¹

From the fitted tensor, DTI derives rotationally invariant scalars—MD and FA—that do not depend on how the patient is oriented in the magnet, plus the principal diffusion direction used to reconstruct fiber tracts. ^{1, 2} This guide explains that math, the acquisition choices that make the tensor estimable, and the QC that keeps the derived numbers trustworthy. For the single-ADC foundation this builds on, see DWI and quantitative ADC QC.

Topic Explanation

From a scalar to a tensor

In free water, diffusion is isotropic and described by a single number. In anisotropic tissue, diffusion must be described by a diffusion tensor $\mathbf{D}$—a symmetric 3×3 matrix with six independent elements:

$$\mathbf{D}=\left(\begin{array}{ccc}{D}_{xx}& D_{xy}& D_{xz}\\ D_{xy}& D_{yy}& D_{yz}\\ D_{xz}& D_{yz}& D_{zz}\end{array}\right)$$

Because the tensor has six independent components, it cannot be solved from one or three diffusion measurements. DTI acquires diffusion weighting along at least six non-collinear directions, plus one non-diffusion-weighted (b=0) image, and fits the six tensor elements voxel by voxel. ¹

Eigenvalues and the diffusion ellipsoid

Diagonalizing the tensor yields three eigenvalues (${\lambda }_1\ge {\lambda }_2\ge {\lambda }_3$), the diffusivities along three perpendicular principal axes, and three eigenvectors that define those axes. ¹ The result is a diffusion ellipsoid: a sphere in isotropic tissue, a cigar shape in a coherent fiber tract where ${\lambda }_1\gg {\lambda }_2\approx {\lambda }_3$. The principal eigenvector (${\lambda }_1$'s direction) points along the dominant fiber orientation and is the basis of tractography.

The scalar maps: MD and FA

Two rotationally invariant scalars summarize the tensor. Mean diffusivity is the average of the eigenvalues:

$$\text{MD}=\frac{{\lambda }_1+{\lambda }_2+{\lambda }_3}{3}=\frac{\text{Trace}(\mathbf{D})}{3}$$

Fractional anisotropy quantifies how unequal the eigenvalues are—how directional the diffusion is:

$$\text{FA}=\sqrt{\frac{3}{2}}\cdot \frac{\sqrt{({\lambda }_1-\text{MD}{)}^2+({\lambda }_2-\text{MD}{)}^2+({\lambda }_3-\text{MD}{)}^2}}{\sqrt{{\lambda }_1^2+{\lambda }_2^2+{\lambda }_3^2}}$$

FA ranges from 0 (isotropic—all eigenvalues equal) to 1 (all diffusion along one axis). ² In healthy brain, the trace of the tensor is fairly uniform at about 2100 × 10⁻⁶ mm²/s (MD ≈ 0.7 × 10⁻³ mm²/s), while FA varies widely by region: high in the corpus callosum and pyramidal tracts, low where fibers cross or fan. ³ In coherent white matter, diffusivity parallel to the fibers (≈ 1400–1800 × 10⁻⁶ mm²/s) is nearly ten times the perpendicular diffusivity (≈ 150–300 × 10⁻⁶ mm²/s). ²

Key Technical Principles

The b-value and the Stejskal-Tanner relation

The strength of diffusion weighting is the b-value, set by the gradient amplitude and timing. For a Stejskal-Tanner pulsed-gradient spin-echo with gradient amplitude $G$, pulse duration $\delta$, and separation $\Delta$:

$$b={\gamma }^2{G}^2{\delta }^2\left(\Delta -\frac{\delta }{3}\right)$$

where $\gamma$ is the gyromagnetic ratio. The diffusion-weighted signal falls off exponentially with b and the diffusion along the gradient direction $\hat{g}$:

$$S=S_0{e}^{-b{\hat{g}}^{\mathsf{T}}\mathbf{D}\hat{g}}$$

Acquiring $S$ along at least six directions and one $S_0$ lets the six tensor elements be fit. Clinical brain DTI typically uses a b-value near 1000 s/mm²: high enough for diffusion contrast, low enough to retain SNR. ¹

Worked example: computing FA

Suppose a white-matter voxel has eigenvalues (in units of 10⁻³ mm²/s): ${\lambda }_1=1.70$, ${\lambda }_2=0.30$, ${\lambda }_3=0.30$. First the mean diffusivity:

$$\text{MD}=\frac{1.70+0.30+0.30}{3}=0.767$$

The numerator of FA uses the squared deviations:

$$(1.70-0.767{)}^2+(0.30-0.767{)}^2+(0.30-0.767{)}^2=0.870+0.218+0.218=1.306$$

The denominator uses the squared eigenvalues:

$${\lambda }_1^2+{\lambda }_2^2+{\lambda }_3^2=2.89+0.09+0.09=3.07$$

So:

$$\text{FA}=\sqrt{\frac{3}{2}\cdot \frac{1.306}{3.07}}=\sqrt{1.5\times 0.4254}=\sqrt{0.638}\approx 0.80$$

An FA of about 0.80 is consistent with a highly coherent white-matter tract such as the corpus callosum. The units of 10⁻³ mm²/s cancel, so FA is dimensionless—one reason it is comparable across scanners when acquisition is controlled.

DWI, DTI, and DKI compared

DTI sits in a family of diffusion models of increasing complexity.

DWI answers "how fast does water diffuse here?"; DTI adds "in which direction, and how coherently?" ^{1, 2}

Why SNR dominates DTI quality

DTI metrics are exquisitely sensitive to noise. Because FA is computed from differences between eigenvalues, random noise makes an isotropic voxel's eigenvalues differ by chance, producing a positive FA bias that grows as SNR falls. Published work confirms FA is biased upward at low SNR while MD stays comparatively unbiased, and that accurate FA—especially in low-FA gray matter—requires substantially more SNR than white matter. ⁴ This single fact shapes the whole QC program: if the b=0 SNR is inadequate, every FA number downstream is suspect.

Clinical Impact

What DTI adds clinically

DTI has moved from research into routine protocols. ¹ Its FA and MD maps reveal subtle white-matter abnormalities in conditions ranging from stroke and multiple sclerosis to traumatic brain injury, and its tractography shows the spatial relationship between a tumor and critical fiber tracts for neurosurgical and radiation-therapy planning. The quantitative nature is the point: a change in FA along a tract can be followed over time or compared between groups—but only if the measurement is reproducible.

Reproducibility depends on acquisition control

The clinical and research value of a FA or MD number rests on its reproducibility across sessions and scanners. That reproducibility is a direct function of SNR, the number and distribution of gradient directions, b-value accuracy, and distortion control. ⁴ A DTI result reported without attention to these is a number without an error bar. For how the gradient subsystem underpins accurate diffusion encoding, see gradient and eddy-current QC.

The role of standardized profiles

Quantitative diffusion imaging is the subject of formal standardization efforts. The RSNA Quantitative Imaging Biomarkers Alliance (QIBA) Diffusion-Weighted MRI Profile defines acquisition and QC requirements—including temperature-controlled phantom measurements—so that diffusion metrics can serve as biomarkers with known bias and reproducibility limits. ⁸ Aligning a DTI program with such a profile is what separates a trustworthy quantitative result from a pretty map.

Practical Optimization Tips

A dependable DTI program is built on controlling the handful of factors that drive the numbers.

1. Secure enough SNR

Because FA bias is driven by low SNR, confirm adequate b=0 SNR for the coil, field strength, and resolution before trusting FA maps—especially in gray matter. ⁴ Resist pushing resolution so high that SNR collapses.

2. Use enough well-distributed directions

Six directions is the mathematical minimum; 20–30 or more evenly distributed directions give more robust, less orientation-biased FA and MD. ¹ Favor uniformly sampled direction schemes.

3. Choose the b-value deliberately

A b-value near 1000 s/mm² balances diffusion contrast against SNR for brain DTI. ¹ Higher b-values increase contrast but cost signal; verify the b-value is accurate, not just nominal.

4. Correct distortions

Echo-planar DTI is prone to eddy-current, susceptibility, and motion distortions; apply vendor and post-processing corrections and verify geometric fidelity. ⁶ Untreated distortion misregisters directions and corrupts tractography.

5. Run a quantitative phantom check

Scan an isotropic phantom and confirm FA near zero and ADC consistent with the reference value at the measured temperature. A nonzero phantom FA is a direct flag of SNR or system problems.

A representative DTI QC checklist

Common pitfalls to avoid

• Trusting FA at low SNR. Noise inflates FA; an isotropic phantom should read near zero, and gray-matter FA needs high SNR. ⁴
• Using the bare six-direction minimum for quantitative work. More, well-distributed directions improve robustness. ¹
• Ignoring temperature in phantom ADC. Water ADC changes sharply with temperature; a reading without a temperature is uninterpretable.
• Skipping distortion correction. Eddy-current and susceptibility distortions misregister directions and degrade tractography. ⁶
• Assuming "nominal" equals "accurate." Verify b-value and gradient directions rather than trusting the prescription.

Regulatory Considerations

DTI quality control fits within the broader MRI quality and accreditation framework rather than a single rule dedicated to diffusion imaging. The program should be documented and defensible, with quantitative diffusion added on top of routine MRI QC.

Key references and expectations:

• ACR MRI Quality Control Manual (2015 edition) — the current manual that ACR-accredited and applying facilities must follow, defining the routine QC program (including SNR, geometric accuracy, and artifact assessment) on which diffusion QC is layered. ⁷
• QIBA Diffusion-Weighted MRI Profile — consensus acquisition and QC requirements for quantitative diffusion metrics, including temperature-controlled phantom measurement. ⁸
• ACR MRI accreditation — requires a documented QC program, qualified medical-physicist or MR-scientist oversight, and an annual equipment performance evaluation. ⁷
• Joint Commission diagnostic imaging standards — require an annual MRI equipment performance evaluation by a qualified medical physicist or MR scientist and ongoing QC. ⁹

Because MRI is non-ionizing, DTI does not fall under NRC or state radioactive-material or x-ray-machine regulation. The binding requirements come from accreditation bodies (ACR) and, where applicable, the Joint Commission, not from a radiation-control program. Of the states DRPS serves, this distinction matters: MRI quality is governed by accreditation and equipment-performance standards rather than the x-ray or radioactive-material rules that cover other modalities. For the accreditation-phantom foundation, see the ACR MRI phantom QC.

Frequently Asked Questions (FAQs)

What is the difference between DWI and DTI?

Diffusion-weighted imaging (DWI) measures how fast water diffuses and summarizes it as a single apparent diffusion coefficient (ADC) per voxel, treating diffusion as the same in all directions. Diffusion tensor imaging (DTI) measures diffusion along at least six non-collinear directions and fits a 3x3 diffusion tensor, which captures both the magnitude and the directionality of diffusion. From the tensor, DTI derives mean diffusivity (direction-independent, similar to ADC) and fractional anisotropy (how directional the diffusion is), plus the principal fiber direction used for tractography.

How many diffusion directions does DTI need?

The diffusion tensor has six independent elements, so a minimum of six non-collinear diffusion-encoding directions plus one non-diffusion-weighted (b=0) image is mathematically required to solve for it. In practice, robust and reproducible fractional anisotropy and mean diffusivity estimates use more—commonly 20 to 30 or more directions—because additional, well-distributed directions reduce the orientational bias and noise sensitivity of the derived metrics.

What is fractional anisotropy and what values are normal?

Fractional anisotropy (FA) is a scalar between 0 and 1 that describes how directional water diffusion is in a voxel. FA near 0 means isotropic diffusion (equal in all directions, as in free water or gray matter), and FA approaching 1 means highly directional diffusion, as in tightly packed, coherently oriented white-matter tracts. Highly organized tracts such as the corpus callosum show high FA, while regions where fibers cross or fan show lower FA.

Why does signal-to-noise ratio matter so much in DTI quality control?

Fractional anisotropy is biased upward at low signal-to-noise ratio: noise makes isotropic regions appear falsely anisotropic. Published work shows FA acquires a positive bias as SNR decreases, while mean diffusivity is comparatively robust. Because of this, a DTI QC program must verify b=0 SNR and confirm that an isotropic water phantom reads FA near zero—an elevated phantom FA is a direct signature of inadequate SNR or a system problem.

What should an isotropic water phantom read for ADC and FA?

In an isotropic liquid phantom, fractional anisotropy should be close to zero because diffusion is the same in all directions; a clearly nonzero value flags a problem. The apparent diffusion coefficient of pure water is strongly temperature dependent—about 1.1 x 10^-3 mm^2/s at 0 degrees C (the ice-water condition used for diffusion calibration), roughly 2.0 x 10^-3 mm^2/s near 20 degrees C, and about 3.0 x 10^-3 mm^2/s at body temperature. Because of this sensitivity, phantom ADC must always be interpreted against the measured temperature.

What are the main sources of error in DTI?

Common pitfalls include eddy-current and echo-planar distortions, patient motion, magnetic-susceptibility artifacts at tissue-air interfaces, gradient nonlinearity that biases the b-value spatially, and low SNR that inflates FA. Each step from acquisition through tensor fitting and region-of-interest analysis can introduce bias, which is why a documented QC program and careful post-processing are essential for trustworthy DTI metrics.

Can a medical physicist set up a DTI quality-control program?

Yes. A qualified or board-certified medical physicist establishes b=0 SNR baselines, verifies ADC accuracy and near-zero FA in an isotropic phantom at a controlled temperature, checks b-value and gradient-direction accuracy and geometric distortion, and documents the program against ACR MRI QC and quantitative-imaging guidance. DRPS provides MRI acceptance testing, annual physics surveys, and quantitative-imaging QC as part of its diagnostic medical physics services.

Key Takeaways

• DTI models diffusion as a tensor. Six independent tensor elements require at least six non-collinear directions plus a b=0 image; robust metrics use 20–30 or more.
• FA and MD come from eigenvalues. MD is the average diffusivity; FA (0–1) measures how directional diffusion is, high in coherent white matter.
• The b-value sets diffusion weighting. Brain DTI typically uses ~1000 s/mm², and the Stejskal-Tanner relation ties b to gradient amplitude and timing.
• SNR drives DTI QC. FA is biased upward at low SNR; an isotropic phantom should read FA near zero.
• Water ADC is temperature dependent. From ~1.1 × 10⁻³ mm²/s at 0 °C to ~3.0 × 10⁻³ at body temperature—so phantom ADC needs a logged temperature.
• Standards exist. The ACR MRI QC Manual and the QIBA DWI Profile define the QC that makes diffusion metrics trustworthy.

Conclusion

DTI's power is that it measures tissue microstructure rather than merely picturing it—resolving the direction and coherence of water diffusion that a single ADC cannot. That power comes with an obligation: the FA and MD values are only as good as the SNR, gradient directions, b-value accuracy, and distortion control behind them.

A dependable DTI program treats those factors as a quality system, not an afterthought. Secure adequate SNR, use enough well-distributed directions, verify the b-value, correct distortions, and confirm with an isotropic phantom at a known temperature that FA reads near zero and ADC matches reference. Done this way—and documented against ACR MRI QC and the QIBA profile—DTI delivers numbers a clinician or researcher can actually trust.

How DRPS Can Help

Diagnostic Radiation Physics Services helps MRI programs turn quantitative diffusion imaging into a controlled, documented measurement. This can include MRI acceptance testing and annual physics surveys, b=0 SNR and geometric-fidelity baselines, isotropic-phantom FA and temperature-referenced ADC checks, b-value and gradient-direction verification, accreditation support, and medical physicist consulting for quantitative-imaging protocols.

DRPS supports facilities across our service locations, including Florida, Maryland, Virginia, Washington DC, California, Nevada, New York, Pennsylvania, New Jersey, and Delaware.

A diffusion map is only a biomarker when the number under the color has a known error bar—and that is what a physics QC program delivers.

Related Resources

• DWI and quantitative ADC QC
• ACR MRI phantom QC
• MRI SNR and RF coil QC
• Gradient and eddy-current QC
• MRI geometric distortion QC
• MRI relaxometry (T1/T2 mapping) QC
• MRI physics testing
• Accreditation support

References

1. Basser PJ, Mattiello J, LeBihan D. MR diffusion tensor spectroscopy and imaging. Biophysical Journal. 1994;66(1):259-267. doi:10.1016/S0006-3495(94)80775-1. PubMed
2. Pierpaoli C, Basser PJ. Toward a quantitative assessment of diffusion anisotropy. Magnetic Resonance in Medicine. 1996;36(6):893-906. doi:10.1002/mrm.1910360612. PubMed
3. Pierpaoli C, Jezzard P, Basser PJ, Barnett A, Di Chiro G. Diffusion tensor MR imaging of the human brain. Radiology. 1996;201(3):637-648. doi:10.1148/radiology.201.3.8939209. PubMed
4. Farrell JAD, Landman BA, Jones CK, et al. Effects of signal-to-noise ratio on the accuracy and reproducibility of diffusion tensor imaging-derived fractional anisotropy, mean diffusivity, and principal eigenvector measurements at 1.5 T. Journal of Magnetic Resonance Imaging. 2007;26(3):756-767. doi:10.1002/jmri.21053. PubMed
5. Le Bihan D, Mangin JF, Poupon C, et al. Diffusion tensor imaging: concepts and applications. Journal of Magnetic Resonance Imaging. 2001;13(4):534-546. doi:10.1002/jmri.1076. PubMed
6. Jones DK, Cercignani M. Twenty-five pitfalls in the analysis of diffusion MRI data. NMR in Biomedicine. 2010;23(7):803-820. doi:10.1002/nbm.1543. PubMed
7. American College of Radiology. Magnetic Resonance Imaging Quality Control Manual. 2015. acr.org
8. Radiological Society of North America, Quantitative Imaging Biomarkers Alliance. QIBA Profile: Diffusion-Weighted Magnetic Resonance Imaging (DWI). qibawiki.rsna.org
9. Stejskal EO, Tanner JE. Spin diffusion measurements: spin echoes in the presence of a time-dependent field gradient. Journal of Chemical Physics. 1965;42(1):288-292. doi:10.1063/1.1695690. doi.org