Transformer-based dual-domain network improves sparse-view CT reconstruction

Cross-Domain TransNet, a dual-domain Transformer, enhanced sparse-view CT reconstruction quality on the NIH-AAPM dataset, suppressing noise/artifacts and outperforming conventional and deep learning methods.

• Cross-Domain TransNet combines image and sinogram domains with a hybrid self-attention mechanism and a Convolution Fusion Layer.
• On the NIH-AAPM dataset, it consistently improved reconstruction quality, suppressed noise, and reduced artifacts under various sparse-view settings.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.