Raw data consistent deep learning-based field of view extension for dual-source dual-energy CT

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Dual-source dual-energy CT (DECT) systems face technical limitations, restricting the field of measurement (FOM) of one detector pair to approximately 35 cm.
• This limited FOM restricts DECT applications to smaller patients, excluding larger individuals from comprehensive spectral analysis.
• Missing spectral information in the periphery of the patient scan hinders the full utility of DECT for diverse patient populations.

Purpose of the Study:

• To develop and evaluate a deep learning-based iterative reconstruction technique to recover spectral information outside the limited FOM of DECT.
• To enable dual-energy applications for the entire patient cross-section, overcoming current FOM constraints.
• To improve the diagnostic capabilities of DECT for larger patients by restoring missing data.

Main Methods:

• A deep learning-based iterative reconstruction algorithm was proposed, utilizing a neural network to refine CT estimates.
• The algorithm uses the reconstruction from the larger FOM (50 cm) as an initial estimate and refines it iteratively.
• Training involved simulated chest, abdomen, and pelvis scans derived from 70 full-body CT datasets, with validation on simulated and measured DECT scans.

Main Results:

• The proposed deep learning approach successfully generated artifact-free CT reconstructions for the entire patient cross-section, including areas outside the limited FOM.
• Simulated data showed an average reconstruction error of 10–17 HU, approximately half that of reference methods.
• Real phantom measurements achieved a similar performance with an average error of 8 HU.

Conclusions:

• The deep learning-based iterative reconstruction effectively recovers missing dual-energy information in dual-source CT systems.
• This method enables dual-energy applications to encompass the entire patient cross-section, regardless of FOM limitations.
• The technique holds significant potential for expanding the clinical utility of DECT across a broader range of patient sizes.