Limited-view CT Reconstruction Based on Autoencoder-like Generative Adversarial Networks with Joint Loss

Area of Science:

• Medical Imaging
• Radiology
• Computer Vision

Background:

• Limited-view X-ray computed tomography (CT) reduces radiation dose but degrades image quality.
• Ionizing radiation damage is a concern in medical imaging.
• Reconstructing high-quality CT images from limited views remains a challenge.

Purpose of the Study:

• To develop a method for predicting missing projections in limited-view CT.
• To improve the quality of reconstructed CT images using a generative approach.
• To reduce radiation dose in CT scans without compromising image fidelity.

Main Methods:

• An autoencoder-like generative adversarial network (GAN) was constructed.
• A convolutional neural network (CNN) within the generator predicts missing sinogram data.
• A joint loss function, incorporating reconstruction and adversarial losses, was employed.
• Filtered back-projection (FBP) reconstructs CT images from completed projections.

Main Results:

• The proposed GAN effectively generates realistic missing projections.
• The joint loss function enhances the sharpness and detail of reconstructed images.
• The method significantly improves limited-view CT reconstruction quality.
• Achieved considerable improvement in limited-view CT reconstruction.

Conclusions:

• The developed GAN-based method successfully addresses the challenge of limited-view CT reconstruction.
• This approach offers a promising solution for dose reduction in CT while maintaining diagnostic image quality.
• The technique has the potential to enhance patient safety in radiological procedures.