A novel fast kilovoltage switching dual-energy computed tomography technique with deep learning: Utility for non-invasive assessments of liver fibrosis

Abstract

Purpose: To investigate whether the iodine density of liver parenchyma in the equilibrium phase and extracellular volume fraction (ECV) measured by deep learning-based spectral computed tomography (CT) can enable noninvasive liver fibrosis staging.

Method: We retrospectively analyzed 63 patients who underwent dynamic CT using deep learning-based spectral CT before a hepatectomy or liver transplantation. The iodine densities of the liver parenchyma (I-liver) and abdominal aorta (I-aorta) were independently measured by two radiologists using iodine density images at the equilibrium phase. The iodine-density ratio (I-ratio: I-liver/I-aorta) and CT-ECV were calculated. Spearman's rank correlation analysis was used to evaluate the relationship between the I-ratio or CT-ECV and liver fibrosis stage, and receiver operating characteristic (ROC) analysis was used to evaluate the diagnostic performances of the I-ratio and CT-ECV.

Results: The I-ratio and CT-ECV showed significant positive correlations with liver fibrosis stage (ρ = 0.648, p < 0.0001 and ρ = 0.723, p < 0.0001, respectively). The areas under the ROC curve for the CT-ECV were 0.882 (F0 vs ≥ F1), 0.873 (≤F1 vs ≥ F2), 0.848 (≤F2 vs ≥ F3), and 0.891 (≤F3 vs F4).

Conclusions: Deep learning-based spectral CT may be useful for noninvasive assessments of liver fibrosis.

Keywords: Deep learning-based spectral CT; Extracellular volume; Iodine density; Liver fibrosis.