An effective sinogram inpainting for complementary limited-angle dual-energy computed tomography imaging using...

Area of Science:

• Medical Imaging
• Radiology
• Computer Vision

Background:

• Dual-energy computed tomography (DECT) offers enhanced diagnostic information but typically requires full 360° scans.
• Conventional DECT systems necessitate high radiation doses and large spatial configurations.
• Limited-angle scanning in DECT presents reconstruction challenges due to data truncation and singularities.

Purpose of the Study:

• To introduce a novel DECT using complementary limited-angle scan (DECT-CL) technology.
• To develop an advanced deep learning method for reconstructing high-quality DECT images from limited-angle data.
• To reduce radiation dose and compress imaging system size while maintaining diagnostic accuracy.

Main Methods:

• A DECT-CL approach with a 180° total scan (90° low-energy, 90° high-energy).
• Development of a complementary-sinogram-inpainting generative adversarial network (CSI-GAN) with a sinogram loss function.
• Image reconstruction using simultaneous algebraic reconstruction technique with total variation (SART-TV).

Main Results:

• The CSI-GAN method effectively inpaints truncated sinograms, suppressing artifacts.
• DECT-CL with CSI-GAN demonstrated superior image reconstruction compared to traditional algorithms.
• Quantitative analysis showed an average 40% RMSE reduction and 26% PSNR increase compared to Patch-GAN.

Conclusions:

• The proposed DECT-CL method effectively addresses the complementary limited-angle problem.
• CSI-GAN significantly improves artifact suppression and image quality in low-dose DECT.
• This technology holds promise for reduced radiation exposure and more compact DECT systems.