Hierarchical decomposed dual-domain deep learning for sparse-view CT reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Deep Learning

Background:

• Sparse-view X-ray computed tomography (CT) reduces radiation dose but causes streaking artifacts with analytic reconstruction.
• Deep learning (DL) methods show promise in artifact reduction but lack theoretical justification for sparse-view CT.

Purpose of the Study:

• To develop a theoretically justified dual-domain deep learning framework for sparse-view CT reconstruction.
• To address the limitations of conventional image-domain and projection-domain DL methods in sparse-view CT.

Main Methods:

• Leveraged deep convolutional framelets (DCF) theory and hierarchical measurement decomposition.
• Proposed a novel dual-domain DL framework utilizing hierarchical decomposed measurements.
• Enhanced projection-domain network performance using DCF's low-rank property and Fourier domain bowtie support.

Main Results:

• Demonstrated performance improvement of the proposed dual-domain DL framework.
• Achieved superior reconstruction performance compared to conventional analytic and DL methods.
• The framework's effectiveness is attributed to the low-rank property of DCF.

Conclusions:

• The study provides a theoretically justified DL approach for sparse-view CT reconstruction.
• The dual-domain DL framework offers a superior alternative for high-quality, low-dose CT imaging.
• Opens new research avenues in medical imaging and advances safer diagnostic techniques.