Hybrid Reconstruction Approach for Polychromatic Computed Tomography in Highly Limited-Data Scenarios

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Medicine

Background:

• Beam-hardening artifacts are a significant challenge in computed tomography (CT), particularly in low-dose or limited-data acquisition scenarios.
• Existing mitigation strategies, such as postprocessing and iterative reconstruction, have limitations in computational cost or effectiveness under data constraints.
• Deep learning (DL) methods show promise for limited-data CT, but their application to beam-hardening artifacts, especially with random projections and limited angular span, remains underexplored.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based framework for mitigating beam-hardening artifacts in limited-data computed tomography (CT).
• To address the challenges posed by randomly selected projections and a highly limited angular span in CT imaging.
• To improve the quality of CT images in scenarios where conventional methods are insufficient.

Main Methods:

• Proposed a deep learning-based prior image constrained (PICDL) framework, a hybrid approach combining a modified Prior Image Constrained Compressed Sensing (PICCS) algorithm (L2-PICCS) with a DL-generated prior image.
• The DL model utilizes a modified U-Net architecture, incorporating ResNet-34 in the encoder for enhanced feature extraction.
• Evaluated the method using rodent head studies on a small-animal CT scanner.

Main Results:

• The PICDL framework successfully corrected beam-hardening artifacts in limited-data CT scenarios.
• The method recovered patient contours and compensated for streak and deformation artifacts, even with a limited angular span and randomly selected projections.
• Hallucinations in the DL-generated prior image were eliminated by the L2-PICCS algorithm, while essential target information was preserved.

Conclusions:

• The proposed PICDL framework offers an effective solution for beam-hardening artifact reduction in challenging limited-data CT imaging.
• This hybrid DL approach demonstrates superior performance compared to conventional methods in scenarios with severe data constraints.
• The study highlights the potential of DL-integrated iterative reconstruction techniques for advancing medical imaging quality and diagnostic accuracy.