Contrast-guided Virtual Monoenergetic Image Synthesis via Adversarial Learning for Coronary CT Angiography using...

Area of Science:

• Medical imaging
• Artificial intelligence
• Cardiovascular diagnostics

Background:

• Coronary CT angiography (cCTA) for coronary artery disease (CAD) diagnosis is limited by blooming artifacts from calcifications and stents.
• Photon counting detector CT (PCD-CT) offers virtual monoenergetic images (VMIs) with a trade-off between artifact reduction and contrast enhancement at different keV levels.

Purpose of the Study:

• To introduce CITRINE, a contrast-guided virtual monoenergetic image synthesis framework using adversarial learning.
• To integrate beneficial spectral characteristics from various keV levels for improved cCTA image quality.

Main Methods:

• CITRINE was trained and validated using cardiac PCD-CT images, specifically 70 keV and 100 keV VMIs.
• The framework utilizes adversarial learning to synthesize images combining low-artifact and high-contrast features.

Main Results:

• CITRINE synthesized images demonstrated reduced blooming artifacts (similar to 100 keV VMI) and high coronary lumen iodine contrast (comparable to 70 keV VMI).
• Quantitative and qualitative evaluations showed improved image quality and reduced percent diameter stenosis by approximately 25% compared to 70 keV VMI.

Conclusions:

• CITRINE effectively synthesizes cCTA images by merging advantages from different keV VMIs.
• This technology enhances diagnostic accuracy and efficiency in cCTA by optimizing PCD-CT data utilization.