Using a single abdominal computed tomography image to differentiate five contrast-enhancement phases: A...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Quantitative imaging biomarkers (radiomics) require high-quality contrast-enhancement in medical images.
• Clinical adoption of radiomics necessitates robust quality control measures for CT scans.

Purpose of the Study:

• To develop a machine-learning algorithm for automated Quality Control of Contrast-Enhancement on CT-scan (CECT-QC).
• To identify five distinct contrast-enhancement phases using CT scan data.

Main Methods:

• A random forest classifier was trained on multicenter CT data from four cohorts (n=503 patients).
• The algorithm used mean intensity of the aorta and portal vein from a single CT image to predict enhancement phases.
• Performance was evaluated on training and test sets, with clinical utility assessed across multiple cohorts.

Main Results:

• The CECT-QC algorithm achieved high accuracy: 98% for non-contrast, 90% for optimal-arterial, and 84% for optimal-portal phases.
• Optimal-portal phase was achieved in 50% of patients and correlated with peak liver malignancy density.
• Contrast-enhancement quality significantly impacted radiomics features used to decipher liver neoplasm phenotypes.

Conclusions:

• A single CT image is sufficient to differentiate five contrast-enhancement phases.
• The CECT-QC algorithm supports radiomics-based precision medicine for common liver neoplasms.
• This approach is applicable to patients with or without liver cirrhosis.