Work Description

Title: Quantification of hepatic steatosis on contrast-enhanced computed tomography scans using artificial intelligence tools [Dataset] Embargo Deposited

h
Embargo release date
  • 11/16/2025
Attribute Value
Methodology
  • Scan phase was identified as one of four categories (non-contrast, arterial, venous, or delayed) using a ResNet-50 machine learning classification model. Automated volumetric segmentation of liver and spleen was performed using Total Segmentator `total' model version 2.4.0. Manual circular regions of interest (ROI) were placed in the liver and spleen of all CT scan phases of a randomly selected subset of all study cohort subjects. Following prior guidance, eight ROIs were placed in the liver (two ROIs in each of liver lobes V, VI, VII, VIII) and three ROIs in the spleen (one ROI in each of the lower, middle, upper third). Mean attenuation was calculated from all pixels across all ROIs within liver ($L_{roi}$) and spleen ($S_{roi}$), and from automated liver ($L_{vol}$) and spleen ($S_{vol}$) volumes. Liver-spleen difference (`L-S') was calculated as $L-S = (L_{vol} - S_{vol})$, and liver/spleen ratio (`L/S') was calculated as $L/S = (L_{vol} - S_{vol})$. Moderate to severe hepatic steatosis was defined for each individual based on non-contrast liver threshold <40 HU, the reference standard. Non-linear regression was used to find the optimal coefficients for three phase-specific (arterial, venous, delayed) increasing exponential decay equations relating contrast-enhanced mean liver attenuation (L') to non-contrast L: $L = \alpha+\beta*(1-e^{-L'*\kappa})$. Equivalent MRI-PDFF (FF) was calculated using the equation: $FF = -.58*L + 38.2$ using the non-contrast L and the corrected contrast-enhanced values of L.
Description
  • The derivation cohort consisted of 1740 subjects drawn from two distinct sub-cohorts at the University of Michigan: (1) a healthy reference population of adult kidney donor candidates (n=1555) that had been previously studied, and (2) an opportunistic clinical cohort with multi-phase CT acquisitions performed in a single examination (n=185). CT scans of the latter cohort were performed for issues related to hematuria (n=114) and for diseases of the adrenals (n=18), liver (n=13), kidneys (n=13), and pancreas (n=2), and other reasons (n=25). Patients were included if they were over 18 years of age; had CT scans performed with a GE Discovery or Lightspeed scanner, using the GE Standard convolution kernel at 120 kVp and slice thickness of 0.625, 1.25, 2.5, or 5 mm, with non-contrast series plus at least one contrast-enhanced phase (arterial, venous, delayed) imaging series in the same examination, with the full body torso visible in the axial field of view.

  • An external validation cohort was used to assess performance of the correction equations. This cohort had the same inclusion criteria of multi-phase CT acquisitions performed in a single examination (n=1044 subjects) and was drawn from a previously detailed cohort of Veterans Administration (VA) patients. CTs were performed with GE (Discovery), Philips (Brilliance, iCT, Ingenuity, Precedence), Siemens (Emotion, Sensation, Somatom Definition AS), or Toshiba (Acquilon) scanners, using soft tissue convolution kernels at 80-140 kVp and slice thickness between 1 and 8 mm.
Creator
Creator ORCID iD
Depositor
Depositor creator
  • true
Contact information
Discipline
Keyword
Citations to related material
  • Derstine, B.A. et al. (2025). Steatosis quantification on contrast-enhanced CT scans. ****Forthcoming.****
Resource type
Last modified
  • 06/13/2025
Published
  • 12/10/2024
Language
DOI
  • https://doi.org/10.7302/xkpk-6w96
License

Remediation of Harmful Language

The University of Michigan Library aims to describe its collections in a way that respects the people and communities who create, use, and are represented in them. We encourage you to contact us anonymously if you encounter harmful or problematic language in catalog records or finding aids. More information about our policies and practices is available at Remediation of Harmful Language.