Pseudo dual-energy CT-derived iodine mapping using single-energy CT data based on a convolution neural network.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Computational Imaging

Background:

• Dual-energy CT (DECT) provides valuable material decomposition information but requires specialized scanners.
• Generating DECT-derived images from single-energy CT could expand access to advanced imaging analysis.
• Low-energy CT (CT_low) images are widely available and can serve as a basis for advanced reconstruction.

Purpose of the Study:

• To develop a convolutional neural network (CNN) model for generating pseudo high-energy CT (CT_pseudo_high) images from CT_low images.
• To create pseudo iodine maps (IM_pseudo) and pseudo virtual non-contrast (VNC_pseudo) images for thoracic and abdominal regions.
• To evaluate the accuracy and performance of the generated pseudo DECT images and material-specific images.

Main Methods:

• Eighty patients undergoing DECT examinations were included, with data split for training (55), validation (5), and testing (20).
• A ResUnet model was employed for image generation, trained using CT_low and high-energy CT (CT_high) images.
• Model performance was assessed using CT values, image noise, mean absolute errors (MAEs), and histogram intersections (HIs).

Main Results:

• The CT values of CT_pseudo_high images showed a mean difference of less than 6 HU compared to CT_high images.
• Image noise in CT_pseudo_high was significantly lower than in CT_high images.
• Mean MAEs were below 15 HU, and histogram intersections approached 1.000, indicating high similarity.

Conclusions:

• The developed CNN model successfully generates pseudo DECT images and material-specific images from single-energy CT data.
• This approach enables the creation of advanced imaging products, such as iodine maps and VNC images, using only low-energy CT input.
• The findings suggest a potential for wider application of DECT-derived analysis without requiring dual-energy acquisition.