Estimating dual-energy CT imaging from single-energy CT data with material decomposition convolutional neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Dual-energy computed tomography (DECT) offers valuable material-specific information but faces accessibility challenges due to high costs.
• Standard single-energy CT (SECT) scanners are more prevalent, limiting access to advanced DECT capabilities in many regions.

Purpose of the Study:

• To develop a deep learning model for high-performance DECT imaging using limited data from standard scanners.
• To demonstrate the feasibility and effectiveness of the proposed method in reconstructing DECT images.

Main Methods:

• A deep learning model was trained to leverage energy-domain correlation and anatomical consistency from DECT data.
• The model reconstructed DECT images using fully-sampled low-energy data and single-view high-energy data.
• Validation was performed on two independent cohorts comprising contrast-enhanced and non-contrast spectral CT scans.

Main Results:

• The deep learning approach successfully generated high-performance DECT images.
• The method demonstrated superior performance on DECT applications compared to existing techniques.
• Feasibility was confirmed across diverse datasets, including contrast-enhanced and spectral CT scans.

Conclusions:

• Deep learning can effectively create DECT imaging from limited data, enhancing accessibility.
• This technology has the potential to significantly reduce radiation dose in DECT scans.
• The approach may enable DECT imaging on standard SECT scanners, broadening its clinical utility.