Physics-informed Deep Learning for Dual-Energy Computed Tomography Image Processing.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Single-energy CT (SECT) struggles to differentiate materials with similar attenuation but different compositions.
• Traditional DECT material decomposition algorithms have limitations, resulting in low signal-to-noise ratio (SNR) in material-specific images.

Purpose of the Study:

• To develop a novel framework for reconstructing non-contrast SECT images from DECT scans using a convolutional neural network (CNN).
• To overcome the limitations of conventional physics-based DECT algorithms and improve image fidelity.

Main Methods:

• Training a CNN to leverage the physics of DECT image generation.
• Utilizing anatomic information from training data to enhance image reconstruction.
• Comparing CNN-based reconstruction with traditional physics-based decomposition methods.

Main Results:

• The CNN framework successfully reconstructs non-contrast SECT images from DECT data.
• CNN-based reconstruction demonstrates higher fidelity compared to traditional algorithms.
• The CNN effectively utilizes the full information content of DECT image data.

Conclusions:

• CNNs offer a superior approach for processing DECT images, surpassing traditional methods.
• This AI-driven framework enhances image quality and material differentiation in CT scans.
• The method holds promise for improving diagnostic accuracy in radiology.