Back to MenuPDFF Liver Fat Assistant
Thresholds & reference
| PDFF | Interpretation |
|---|---|
| < 5% | No hepatic steatosis (normal) |
| 5 – <17.5% | Mild steatosis (grade 1) |
| 17.5 – <22% | Moderate steatosis (grade 2) |
| ≥ 22% | Severe steatosis (grade 3) |
Steatosis onset at PDFF >5% (EASL). Severity band edges (≈17.5%, ≈22%) follow Tang et al., Radiology 2013;267:422–431, and are approximate — histologic-grade cutoffs vary by study and are not universally standardized. A proper PDFF sequence corrects for T1 bias, T2*/iron, and fat spectral complexity; low SNR and motion still degrade accuracy. Interpret with clinical context.
The physics & pitfalls (fellow / attending level)
What PDFF actually measures. PDFF is the density of mobile fat protons divided by the density of mobile fat plus water protons. Once the confounders below are removed it is a property of the tissue, not of the scanner, which is why it agrees with MR spectroscopy (mean bias about −0.1 percentage points, R² 0.96) and why field strength, vendor and reconstruction have little effect on it. It is not the histology number. Pathologists grade the percentage of hepatocytes containing fat vacuoles; PDFF counts fat by proton density. The two scales are related but not equal, which is why the PDFF grade cutoffs (5, about 17.5, about 22%) sit well below the histologic ones (5, 33, 66%).
How fat and water are separated. The main fat peak (methylene) resonates about 3.4 ppm from water, roughly 220 Hz at 1.5 T and 440 Hz at 3 T, so fat and water drift in and out of phase every 4.6 ms at 1.5 T (2.3 ms at 3 T). A spoiled gradient echo samples that cycle with at least four, preferably six, closely spaced echoes, and a model of fat plus water plus decay is fitted to how the signal evolves. Fitting the complex data, phase as well as magnitude, tells the fit which component is dominant, so PDFF spans 0 to 100% instead of the 50% ceiling of two-point In-/Out-of-Phase imaging.
What the sequence corrects, and how. T1 bias: fat has a much shorter T1 than water and would be over-weighted, so the flip angle is kept low (for 3D, roughly 3–5° at 1.5 T and 2–4° at 3 T). T2* decay: modeled across the echoes as a single R2*, which also produces an R2* map as a by-product, a free iron screen on every PDFF exam (see the Iron tool). Fat spectral complexity: fat has about six peaks, not one, and a fixed six-peak model accounts for them. Noise bias: at low SNR magnitude data push the estimate away from the truth, which is why large voxels and adequate SNR matter. Eddy-current phase errors: handled in reconstruction, typically by hybrid magnitude and complex fitting.
Field strength is portable here. This is the opposite of the iron tool. R2* scales with field strength and its LIC calibration is 1.5 T only, but a confounder-corrected PDFF is the same number at 1.5 T and 3 T. No conversion is needed between exams on different magnets.
Fat-water swap. The fit has two near-equal solutions, fat-dominant and water-dominant, and field inhomogeneity can push it to the wrong one. A swapped region reads close to 100 minus the true value. It favors the dome, the lateral edges, tissue near metal or bowel gas, and sometimes a whole slice. Look at the PDFF map, not just the ROI number: a segment implausibly higher than its neighbors is a swap until the map says otherwise.
Iron and motion. With heavy iron the later echoes decay into noise and the fat estimate loses precision. The QIBA profile treats PDFF as unreliable above an R2* of about 300 s−1 at 1.5 T or 450 s−1 at 3 T (300 s−1 is an LIC of about 8 mg/g on the Iron tool's calibration). Check the R2* map before trusting a fat fraction in an iron-loaded liver. Breath-hold failure blurs the map and mixes in vessel and edge signal.
ROIs and change over time. The most reproducible method is the largest ROI that fits in each of the nine Couinaud segments, averaged, avoiding vessels, ducts, edges and artifact (Campo et al. 2017); QIBA asks for ROIs of at least 1 cm diameter and prefers 4 cm² or more. Precision: the repeatability coefficient is about 3 percentage points on the same scanner and about 4 across scanners or protocols (Yokoo et al. 2018), and QIBA treats an absolute change of 5 points or more as real with 95% confidence. In MASH treatment trials, a relative drop of 30% or more is associated with histologic response (51% vs 14%; OR about 7) and NASH resolution (Stine et al. 2021). Note the tension at low baseline: 30% of a 10% PDFF is 3 points, inside measurement noise, so compare on the same scanner and protocol and read small changes cautiously.
| Modality | Quant? | Sens | Spec | Mild? | Monitor? | Key threshold |
|---|---|---|---|---|---|---|
| B-mode US | No | 80–89% | 87–90% | Poor | No | Liver > renal cortex echogenicity |
| Hepato-renal index | Semi | ~83% | ~98% | Moderate | Limited | HRI ≥ 1.3 |
| CAP (FibroScan) | Semi | 80–92% | ~90% | Moderate | No | ≥ 275 dB/m (AASLD) |
| US attenuation coeff. | Yes | ~83% | ~77% | Moderate | Under study | ≥ 0.66 dB/cm/MHz |
| Non-contrast CT | Semi | 72% / 82% | 88% / 94% | Poor | No | Liver < 40 HU; L−S < −10 HU |
| Contrast-enhanced CT | Semi | 66% | 90% | Poor | No | Unreliable (see note) |
| DECT fat fraction | Yes | 85% | 88% | Moderate | Under study | FVF ≥ 5.3% |
| In- & Out-of-Phase MRI (2-point Dixon) | Semi | 77–90% | 87–91% | Yes | Limited | Out-of-Phase signal drop |
| MRI-PDFF | Yes | 77–100% | 87–100% | Yes | Yes | PDFF ≥ 5–5.56% |
| MR spectroscopy | Reference | 80–98% | 80–98% | Yes | Yes (research) | Fat fraction ≥ 5.56% |
1 · B-mode ultrasound (grayscale)First-line
Principle
Most widely used first-line test — low cost, no radiation, wide availability. Fat raises hepatic echogenicity.
Criteria
- Increased hepatic echogenicity vs. renal cortex ("bright liver").
- Decreased visibility of intrahepatic vessels and diaphragm; posterior beam attenuation.
- Reliably detectable when >20% of hepatocytes contain lipid.
Hepato-renal index (HRI)
Semi-quantitative liver-to-renal-cortex echogenicity ratio. AUROC 0.91; optimal cutoff HRI ≥ 1.3.
Performance
Sensitivity 80–89%, specificity 87–90% for moderate-severe steatosis; sensitivity falls below 50% for mild steatosis.
Limitations
- Qualitative, subjective, operator-dependent; degraded by obesity.
- Unreliable for mild steatosis and for monitoring therapy.
2 · Controlled attenuation parameter (CAP / FibroScan)
Principle
Point-of-care measure of ultrasound attenuation through the liver, acquired with liver stiffness on vibration-controlled transient elastography.
Thresholds
- ≥ 248 dB/m diagnostic for steatosis; AASLD 2025 uses ≥ 275 dB/m as a convenient screening cutoff.
- 263 dB/m = 90% sensitivity to rule out; 270 dB/m = 100% NPV in low-pretest-probability groups.
- Grades: S0 0–10%, S1 11–33%, S2 34–66%, S3 >67% (AUROC 0.87 / 0.77 / 0.70).
Performance
Sensitivity ~80–90%, specificity ~90% for any steatosis.
Limitations
- Overlap between adjacent grades; not sensitive enough to track treatment response.
- Affected by obesity (XL probe may be needed).
3 · Ultrasound attenuation coefficient (AC / ATT)Emerging
Principle
Newer quantitative tool built into conventional US machines; measures the attenuation coefficient of the beam as an add-on to B-mode.
Thresholds
- ATT ≥ 0.66 dB/cm/MHz: sensitivity 82.7%, specificity 77.3%.
- AC ≥ 0.42 dB/cm/MHz: AUROC 0.82 in one cohort.
Performance
AUROC 0.82–0.89 — comparable to or slightly better than CAP in some studies; scalable on routine machines.
4 · Non-contrast CT (incl. liver–spleen)L−S valid
Principle
Opportunistic detection; fat lowers liver attenuation. Normal liver is ~8–10 HU denser than spleen.
Criteria (validated on non-contrast CT)
- Absolute liver attenuation < 40 HU → ≥ moderate steatosis.
- Liver − spleen difference < −10 HU (liver >10 HU lower than spleen).
- Liver/spleen ratio < 1.0.
Performance
Sensitivity 72% / specificity 88% for any steatosis (≥5% fat); 82% / 94% for ≥ moderate (≥20–33% fat).
Continuous estimate
Beyond thresholds, mean liver attenuation maps to a pathologic fat % (see the Kodama estimator above): fat% = 100 · exp[−exp(−1.915 + 0.051 · HU)] on unenhanced CT.
Limitations
- Poor for mild steatosis; ionizing radiation; not for longitudinal monitoring.
5 · Contrast-enhanced CT & VNCRatio unreliable
Why liver/spleen enhancement ratios fail here
After iodinated contrast, liver and spleen enhance on different time courses — early on the spleen shows heterogeneous arciform ("zebra") enhancement, and the liver–spleen relationship keeps shifting with phase, scan timing, contrast dose, and cardiac output. No single enhancement ratio or difference is reproducible across exams, so a value meaning steatosis at one timing means nothing at another. Absolute and liver/spleen-ratio thresholds are therefore validated only on non-contrast CT. On enhanced scans, use DECT fat fraction or a virtual non-contrast series instead of a ratio.
Performance (enhanced, absolute attenuation)
Sensitivity 66% / specificity 90% for any steatosis — lower than NCCT.
Portal-venous fat equation (historical)
Kodama derived a portal-venous model — fat% = 100 · exp[−exp(−1.781 + 0.024 · HU)] (in the estimator above) — but it is tied to that study's injection protocol and timing, so it does not generalize; treat any enhanced-CT fat number as a rough estimate, not a quantification.
Virtual non-contrast (VNC) from dual-source DECT
Approximates true unenhanced attenuation. VNC 46 HU threshold: 69% sensitivity / 95% specificity for ≥ moderate steatosis (vs. 100% / 42% at the standard 40 HU). Note VNC tends to overestimate true unenhanced HU.
6 · Dual-energy CT (DECT) fat fraction
Principle
Material decomposition generates fat-fraction maps and virtual monoenergetic images; can quantify fat even from contrast-enhanced acquisitions without an extra non-contrast dose.
Thresholds
- DECT fat volume fraction (FVF) ≥ 5.3%; CT value at 74 keV of 55.9 HU for ≥ mild steatosis (sex-specific: 3.8%/56.5 HU male, 2.1%/56.3 HU female).
Performance
Sensitivity 85% / specificity 88%; DECT-FVF correlates strongly with MRI-PDFF (Spearman r = 0.933).
7 · In- and Out-of-Phase (two-point Dixon)Not for quantification
Principle
Water and fat protons precess at different frequencies; signal loss on the Out-of-Phase image relative to the In-Phase image indicates intracellular lipid (Out-of-Phase = opposed-phase). Excellent for detecting and localizing fat.
The two-point fat-fraction equation
The commonly used estimate is FF = (SIIn − SIOut) / (2 × SIIn). It tracks fat at low–moderate levels in a clean liver, but it is a semiquantitative screen, not a true fat fraction — four confounders make it unreliable for quantification:
- Iron / T2* decay. The In-Phase echo is at a longer TE than the Out-of-Phase echo, so iron suppresses In-Phase signal more, shrinking the numerator and underestimating fat; with significant siderosis the In-Phase signal can fall below Out-of-Phase and give paradoxical or negative values. Coexisting steatosis + iron is where it fails hardest.
- 50% ceiling. Magnitude two-point Dixon cannot distinguish 30% from 70% fat — the signal is identical on either side of the fat–water crossover, so it saturates at 50%.
- Multi-peak fat. Fat has ~6 spectral peaks; the single-peak assumption means the signal is never fully out of phase, systematically underestimating fat.
- T1 bias. Without a low flip angle, water is T1-weighted upward and fat fraction is overestimated.
Spleen-normalized “fat %” does not rescue it
Normalizing liver to spleen at each echo (Borra 2009) assumes the two organs share the same T2* — the very assumption iron destroys — so it is arguably more fragile in the patients who most need accuracy.
Appropriate use
Treat the In- and Out-of-Phase result as qualitative-to-semiquantitative (steatosis present/absent, roughly graded) — not as a reportable fat fraction or PDFF.
Reliable alternative
Confounder-corrected multi-echo PDFF (section 8) explicitly models and removes T2*, the multi-peak fat spectrum, T1, and noise — the reference standard for quantification. Multi-point Dixon (3+ echoes) with T2* correction agrees closely with MRS (mean diff −0.5% ± 4.6%).
8 · MRI-PDFFThis tool · reference-grade
Principle
Multi-echo chemical-shift-encoded gradient echo with multipeak spectral fat modeling and T2* correction yields a quantitative fat-fraction map of the whole liver.
Thresholds
- PDFF ≥ 5–5.56% defines steatosis (this calculator uses >5%).
- With multipeak + T2 correction, up to 100% sensitivity/specificity/AUROC for ≥5.56% in one study.
Performance
Sensitivity 77–92%, specificity 87–94% in most series, reaching 100% in the single study with multipeak and T2* correction noted above (hence 77–100% in the comparison table); AUROC 0.96–0.99 vs. histology; correlation with MRS r = 0.95–0.99.
Advantages
- Quantitative, reproducible across vendors/field strengths; whole-liver; BMI-independent (if patient fits); sensitive to treatment change; valid trial endpoint.
Limitations
- Cost, limited availability, MRI-compatibility required; not point-of-care.
9 · MR spectroscopy (¹H-MRS)Reference standard
Principle
Single-voxel proton MRS measures the lipid-to-water signal ratio in a defined liver volume — the non-invasive reference standard used to validate other methods.
Performance
Sensitivity and specificity 80–98%; excellent correlation with histology (r = 0.71–0.96).
Limitations
- Samples a small single voxel (~27 mL), not the whole organ; technically demanding; limited availability; mostly research.
Guideline library
- AASLD 2025 Imaging-Based NILDA Practice Guideline — Sterling RK, et al. Hepatology 2025;81(2):672–724.
- AASLD 2023 NAFLD Practice Guidance — Rinella ME, et al. Hepatology 2023;77(5):1797–1835.
- AHA 2022 Scientific Statement — Duell PB, et al. ATVB 2022;42(6):e168–e185.
- ACR Appropriateness Criteria: Abnormal Liver Function Tests — JACR 2023;20(11S):S302–S314.
- CT meta-analysis — Haghshomar M, et al. Radiology 2024;313(2):e241171.
- CT attenuation–fat equations — Kodama Y, Ng CS, Wu TT, et al. AJR Am J Roentgenol 2007;188(5):1307–1312.
- MRS > CAP — Runge JH, et al. Radiology 2018;286(2):547–556.
- MASLD review — Tilg H, et al. JAMA 2026;335(2):163–174.
- PDFF linearity, bias and precision — Yokoo T, et al. Radiology 2018;286(2):486–498.
- PDFF change and histologic response — Stine JG, et al. Clin Gastroenterol Hepatol 2021;19(11):2274–2283.
- Liver ROI method for PDFF and R2* — Campo CA, et al. AJR 2017;209:592–603.
- QIBA Profile: MRI-based PDFF of the liver — RSNA QIBA, Stage 2 Consensus, 2023.
Educational summary compiled from the above; figures/percentages reflect the cited primary literature and may vary by population and technique.
