Quantification of hepatic steatosis on post-contrast computed tomography scans using artificial intelligence tools

Area of Science:

• Radiology
• Medical Imaging
• Hepatology

Background:

• Early detection of steatotic liver disease (SLD) is crucial for patient outcomes.
• Hepatic steatosis is often incidentally found on computed tomography (CT) scans.
• Existing methods for estimating liver fat fraction (FF) using CT have limitations with post-contrast scans.

Purpose of the Study:

• To determine if an automated workflow can accurately measure liver attenuation on CT.
• To validate thresholds for liver or liver-spleen attenuation in post-contrast CT studies.
• To develop a method for estimating magnetic resonance proton density fat fraction (MR-PDFF) on post-contrast CT.

Main Methods:

• Utilized the TotalSegmentator machine learning model for automated liver and spleen segmentation on CT scans.
• Extracted mean attenuation values from segmented volumes and manual regions of interest (ROIs).
• Developed and validated phase-specific (arterial, venous, delayed) regression equations to estimate MR-PDFF from post-contrast liver attenuation.

Main Results:

• Automated segmentation showed high correlation with manual measurements for liver and spleen attenuation.
• Liver attenuation alone effectively classified moderate-to-severe steatosis on post-contrast CT.
• Estimated fat fraction using corrected post-contrast CT attenuation closely agreed with non-contrast MR-PDFF, achieving high AUROC values.

Conclusions:

• Automated CT analysis provides a reliable alternative to manual ROIs for liver fat assessment.
• Post-contrast CT liver attenuation can identify moderate-to-severe hepatic steatosis.
• Developed correction equations enable MR-PDFF estimation from post-contrast CT, facilitating large-scale SLD screening and research.