Automatic hepatic steatosis quantification using low-dose CT with deep learning-based noise reduction and CT fat...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Hepatic steatosis quantification is crucial for patient management.
• Current CT-derived fat fraction (CDFF) software accuracy may be affected by radiation dose.
• Deep learning (DL) shows promise in improving image quality and quantitative analysis in medical imaging.

Purpose of the Study:

• To evaluate the accuracy of CDFF software for quantifying hepatic steatosis at varying radiation doses.
• To assess the impact of DL-based noise reduction (DLDR) on CDFF accuracy in low-dose CT (LDCT) scans.
• To compare CDFF accuracy against MRI-derived proton density fat fraction (MRI-PDFF) as the reference standard.

Main Methods:

• Retrospective analysis of 125 liver donor candidates with non-contrast CT and MRI data.
• CDFF measurements were performed on full-dose and simulated LDCT scans (50%, 25%, 10% dose).
• DLDR was applied to LDCT scans for CDFF recalculation; accuracy was assessed using Pearson correlation and ROC analysis against MRI-PDFF.

Main Results:

• Full-dose CDFF showed moderate correlation with MRI-PDFF (r=0.728).
• Correlation decreased with lower doses but improved with DLDR application (r=0.725-0.736).
• ROC AUC for hepatic steatosis diagnosis was 0.82 for full-dose CDFF, with sustained performance at reduced doses (except 10%) when using DLDR.

Conclusions:

• CDFF accuracy for hepatic steatosis quantification is dose-dependent, declining at lower radiation levels.
• DLDR significantly enhances CDFF accuracy at reduced radiation doses, maintaining high diagnostic performance.
• DLDR enables potential reduction in patient radiation exposure while ensuring reliable hepatic steatosis assessment.