Deep learning based spectral extrapolation for dual-source, dual-energy x-ray computed tomography.

Area of Science:

• Medical Imaging
• Computed Tomography
• Artificial Intelligence

Background:

• Dual-source, dual-energy CT (DECT) often requires data completion due to limited field of view (FoV).
• Current methods estimate missing data and truncate reconstructions, limiting spectral contrast estimation over larger areas.
• Spectral extrapolation is crucial for advanced applications like model-based iterative reconstruction and contrast-enhanced imaging of large patients.

Purpose of the Study:

• To evaluate the accuracy of spectral extrapolation in DECT.
• To develop and prototype a deep learning algorithm for spectral extrapolation.
• To improve spectral contrast estimation beyond the limitations of standard data completion.

Main Methods:

• A hybrid deep learning model combining a piecewise linear transfer function (PLTF) and a U-net convolutional neural network (CNN) was developed.
• The model was trained on 50 dual-source, dual-energy abdominal CT scans.
• The PLTF mapped spectral contrast, while the CNN refined estimates using structural information and learned contrast relationships.

Main Results:

• The CNN achieved significantly lower extrapolation errors (7.5 HU average) compared to the PLTF (26 HU average).
• The deep learning approach demonstrated robustness and generalization to unseen patient data.
• The integrated system produced high-fidelity spectral CT data for extended FoVs, even with restricted initial data.

Conclusions:

• Deep learning robustly infers spectral contrast from feature-contrast relationships in DECT data.
• The developed method significantly enhances spectral extrapolation performance.
• Future work aims to further refine results for challenging cases by integrating projection data.