Liver Fat Map - v2

PROTOTYPE · Educational & research decision support. Not validated for autonomous diagnosis, staging or treatment selection. Radiologist verification required before PACS dictation

LiverFatMap v3.0

Created by Dr. Sharad Maheshwari MD · Founder, BeResponsibleAI

CT / MRI-PDFF Steatosis Triage

Enter liver attenuation (HU), a measured MRI-PDFF, or liver + spleen HU. One engine converts the input to an estimated PDFF (PDFF % = −0.58 × HU + 38.2, unenhanced 120 kVp) and assigns a category from one selected PDFF cutoff set. Every card, the synthesis, the composite flag and the report read the same result.

Boundaries: imaging-based categories, not histology or measured PDFF. Iron overload and amiodarone raise HU; acute hepatitis/oedema lowers HU; contrast, non-120 kVp acquisition and reconstruction change values. CT precision falls at low fat fractions (<5% PDFF, >~57 HU).

Input basis Liver HU (CT)MRI-PDFF % (measured)Liver + Spleen HU (LAI) PDFF→grade cutoffs

Acquisition preconditions (regression validated only inside this domain):

Unenhanced ~120 kVp No iron / amiodarone / severe artifact

Profound fat (10)Normal (70)

HU (liver)

3-ROI sampler (avoid vessels, ducts, subcapsular margin)

Measured MRI-PDFF (%) Category comes straight from the PDFF cutoffs. No CT conversion is applied.

Spleen HU or L−S difference only Liver HU is entered above. With both organs, PDFF comes from liver HU and LAI is a concordance check. Difference-only gives an uncalibrated impression and no PDFF.

Low-fat uncertainty zone (PDFF ~5–10%): CT–MRI agreement is weakest here and normal livers overlap this range (Boyce 2010 mean 58.8 ± 10.8 HU). Use MRI-PDFF if quantification changes management.

Resolution: outputs change at every 0.1 HU, but CT repeatability is roughly ±3–5 HU. The tool flags any result within 4 HU of a cutoff.

Imaging-based category

S2

Not a biopsy prediction

Estimated PDFF

Not a substitute for MRI-PDFF

Triage status

Step 2 · Optional biomarkers exploratory

Reference norm Unisex (ALT 40, HDL 1.10)Male (ALT 30, HDL 1.03)Female (ALT 19, HDL 1.29)

ALT (U/L) HDL-C (mmol/L)

Zhang et al. (2026) relative odds vs normal baseline:

Derived in bariatric candidates (n=285, pretest MASH ~52%); not externally validated; intercept and calibration not reported. A relative research signal, not an individual probability. ALT and HDL-C never alter the Step 1 category. Citation to be verified.

Composite research signal: category S2 or higher plus abnormal ALT and/or HDL-C. MASH cannot be diagnosed by CT and labs alone.

Synthesis:

Governance notice: the draft separates measurement, interpretation certainty and clinical correlation. Confirm technique, review confounders and edit to the clinical indication.

Report

Technique & confounder attestation: I confirm this is a true unenhanced ~120 kVp acquisition (or MRI-PDFF where entered) and that iron overload, acute hepatitis/oedema and amiodarone have been considered.

Rationale:

Provenance register

Every constant that drives a displayed number. Status reflects what has been checked, not what is likely true. Verify each on PubMed before release.

Constant	Value	Source (as recorded)	Status
CT→PDFF regression	−0.58·HU + 38.2; R²=0.828 (120 kVp, n=72)	Pickhardt PJ et al. AJR 2018;211:W151–W157	VERIFY; residual error not recorded
Cutoff set A	6.4 / 17.4 / 22.1%	Tang A et al. Radiology 2013;267:422–431	VERIFY
Cutoff set B	6.5 / 16.5 / 22.0%	Dioguardi Burgio et al. 2019 (user-supplied)	VERIFY attribution
Cutoff set C	5.75 / 15.5 / 21.35%	Provenance not stated	UNSOURCED
LAI threshold	L−S ≤ −10 HU	Piekarski, Radiology 1980;137:727; Kodama, AJR 2007;188:1307	VERIFY; concordance flag only
ALT norms	M 30, F 19 U/L	Prati D et al. Ann Intern Med 2002;137:1–10	VERIFY
HDL-C cutoffs	M 1.03, F 1.29 mmol/L	Standard metabolic-syndrome cutoffs	VERIFY
Zhang model	OR 0.944/HU, 1.011/ALT, 0.209/HDL	Zhang H et al. BMC Gastroenterol 2026;26:419 (user-supplied)	VERIFY (not confirmed by me)
Boundary flag	< 4 HU from cutoff	Design choice from CT repeatability (±3–5 HU)	Not literature-derived
S2 gate for composite	category ≥ S2 in selected set	Design choice (replaces fixed ≤40 HU)	Single rule, see devil’s advocate

HU boundaries implied by each cutoff set

Computed live: HU = (PDFF − 38.2)/−0.58. A fixed “40 HU” is only one point inside this spread.

Set	S0→S1	S1→S2	S2→S3

Three rulers

Histology counts the fraction of hepatocytes with fat (S0 <5%, S1 5–33%, S2 34–66%, S3 >66%). MRI-PDFF measures the mobile-proton fat fraction, so its numbers run lower than histology (S3 is not >66% PDFF). CT measures X-ray attenuation, which is a surrogate for PDFF and is altered by iron, oedema, beam energy and kernel. They are related instruments, not interchangeable units. Report the measurement and the category as an estimate.

Devil’s advocate

• Finer is not more accurate. Display resolution of 0.1 HU sits inside ±3–5 HU repeatability. A purist would show PDFF with an interval and no category at all.
• Errors compound. CT→PDFF scatter, PDFF→histology scatter and biopsy sampling variability multiply.
• The cutoff set changes the answer. The three sets place the S1/S2 line at about 36–39 HU. Pick one and declare it in the report.
• The S2 gate changes the old behaviour. The previous fixed ≤40 HU gate came from donor and population calibration, not from PDFF cutoffs. Tying the gate to the selected set is more consistent but is a judgement, not a published rule.
• Inverse regression is not X-on-Y. HU shown in PDFF mode would be biased (regression dilution); this version therefore uses PDFF directly and never reports an equivalent HU as a finding.
• LAI may add noise. Spleen attenuation is altered by several conditions; LAI is used here only to flag discordance.
• Opposing view: radiologists need a dictation-ready category, and a purely continuous output may be ignored. Hedged categories are a pragmatic compromise.

References (verify each before release)

1. Pickhardt PJ, Graffy PM, Reeder SB, Hernando D, Li K. Quantification of liver fat content with unenhanced MDCT: phantom and clinical correlation with MRI proton density fat fraction. AJR 2018;211:W151–W157.
2. Tang A, Tan J, Sun M, et al. Nonalcoholic fatty liver disease: MR imaging of liver proton density fat fraction to assess hepatic steatosis. Radiology 2013;267:422–431.
3. Park SH, Kim PN, Kim KW, et al. Macrovesicular hepatic steatosis in living liver donors: use of CT for quantitative and qualitative assessment. Radiology 2006;239:105–112.
4. Boyce CJ, Pickhardt PJ, Kim DH, et al. Hepatic steatosis in asymptomatic adults identified by unenhanced low-dose CT. AJR 2010;194:623–628.
5. Piekarski J, Goldberg HI, Royal SA, et al. Difference between liver and spleen CT numbers in the detection of hepatic fatty infiltration. Radiology 1980;137:727–729.
6. Kodama Y, Ng CS, Wu TT, et al. Comparison of CT methods for determining the fat content of the liver. AJR 2007;188:1307–1312.
7. Prati D, Taioli E, Zanella A, et al. Updated definitions of healthy ranges for serum alanine aminotransferase levels. Ann Intern Med 2002;137:1–10.
8. Frost C, Thompson SG. Correcting for regression dilution bias: comparison of methods for a single predictor variable. J R Stat Soc A 2000;163:173–189.
9. Zhang H, et al. CT-based composite biomarkers for MASH in people with obesity before bariatric surgery. BMC Gastroenterol 2026;26:419. (User-supplied; not independently confirmed.)

The Haghshomar 2024 meta-analysis, the “biopsy correlation cohort” medians and the Borggreve/Rashid lipid citations from v2.1 were removed until their journals, DOIs and findings are verified.