Image synthesis of effective atomic number images using a deep convolutional neural network-based generative...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Materials Science

Background:

• Dual-energy computed tomography (DECT) is crucial for material characterization via effective atomic number (EAN) imaging.
• Synthesizing EAN images from DECT data can improve material decomposition.
• Current methods may require additional scans or complex processing.

Purpose of the Study:

• To develop a novel image synthesis framework for generating EAN images from single-energy CT (SECT) images.
• To utilize a deep convolutional neural network (CNN)-based generative adversarial network (GAN) for this synthesis.
• To assess the accuracy and performance of the synthesized EAN images.

Main Methods:

• A CNN-based GAN was developed to synthesize EAN images from SECT images at 120 kVp.
• The framework was trained and validated using a dataset of CT images.
• Evaluation metrics included Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Mutual Information (MI).

Main Results:

• The synthesized EAN images showed a difference within 9.7% compared to reference EAN values across all regions of interest.
• Average evaluation metrics demonstrated high fidelity: MAE of 0.09, RMSE of 0.045, SSIM of 0.89, PSNR of 54.97, and MI of 1.03.
• The GAN-based approach successfully generated accurate EAN images from SECT data.

Conclusions:

• A robust image synthesis framework was established to derive EAN images from SECT scans.
• This framework enables material decomposition using DECT principles without necessitating extra scans.
• The developed method offers a promising approach for efficient material characterization in CT imaging.