Contrast Media Reduction in Computed Tomography With Deep Learning Using a Generative Adversarial Network in an...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Computed Tomography

Background:

• Reducing iodine-based contrast medium (CM) dose in computed tomography (CT) is crucial for patient safety.
• Generative adversarial networks (GAN) show potential for enhancing low-dose CT images.

Purpose of the Study:

• To evaluate the feasibility of using GAN for virtual contrast enhancement to reduce CM dose in abdominal CT.
• To assess image quality and diagnostic consistency using GAN-generated virtual CM.

Main Methods:

• Multiphasic abdominal low-kilovolt CTs were performed on Göttingen minipigs with low and normal CM doses.
• A GAN was trained for image-to-image conversion from low CM to virtual CM.
• Contrast-to-noise ratio (CNR) was calculated, and a visual Turing test was conducted with radiologists.

Main Results:

• GAN-based virtual contrast enhancement significantly increased vascular and parenchymal CNR compared to low CM dose.
• Virtual CM showed comparable CNR to normal CM dose.
• Radiologists found virtual CM images to be pathologically consistent in 96.5% of cases.

Conclusions:

• GAN-based virtual contrast enhancement is a feasible method to reduce CM dose by approximately 70% in abdominal CT.
• The technique maintains satisfactory image quality and diagnostic consistency.
• This approach holds promise for reducing CM-related risks in CT imaging.