[Reconstruction from CT truncated data based on dual-domain transformer coupled feature learning].

Area of Science:

• Medical imaging
• Artificial intelligence in radiology
• Image reconstruction algorithms

Context:

• Computed tomography (CT) scans often suffer from truncation artifacts due to an insufficient field of view (FOV).
• These artifacts distort image structures and compromise diagnostic accuracy.
• Existing reconstruction methods struggle to fully recover information outside the scanned FOV.

Purpose:

• To introduce DDTrans, a novel CT reconstruction model.
• DDTrans utilizes dual-domain Transformer learning in both projection and image spaces.
• The goal is to minimize truncation artifacts and image distortion.

Summary:

• DDTrans employs Transformer networks in projection and image domains, leveraging attention mechanisms for global feature capture.
• A differentiable Radon back-projection operator enables end-to-end training.
• Projection consistency loss further refines image reconstruction accuracy.

Impact:

• DDTrans demonstrates superior performance in removing edge artifacts and restoring external FOV information compared to existing algorithms.
• The model effectively ensures accurate reconstruction within the FOV.
• It achieves approximate reconstruction of data outside the FOV, enhancing diagnostic utility.