DuDoCFNet: Dual-Domain Coarse-to-Fine Progressive Network for Simultaneous Denoising, Limited-View Reconstruction,...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Nuclear Medicine

Background:

• Low-dose (LD) SPECT reduces radiation but increases noise.
• Limited-view (LV) SPECT accelerates scans but lowers accuracy.
• CT-based attenuation correction (AC) adds radiation and misalignment issues.

Purpose of the Study:

• To develop a multi-task learning method for simultaneous LD denoising, LV reconstruction, and CT-free attenuation map generation in cardiac SPECT.
• To fuse cross-domain and cross-modality information for enhanced accuracy in each task.

Main Methods:

• Proposed a Dual-Domain Coarse-to-Fine Progressive Network (DuDoCFNet).
• Employed paired dual-domain networks with multi-layer fusion for feature integration.
• Utilized two-stage progressive learning in projection and image domains for coarse-to-fine estimations.

Main Results:

• DuDoCFNet demonstrated superior accuracy in projection estimation, μ-map generation, and AC reconstructions.
• Outperformed existing single- and multi-task learning methods across various LD levels.
• Achieved enhanced accuracy by effectively fusing cross-domain and cross-modality information.

Conclusions:

• DuDoCFNet offers a promising solution for simultaneous LD denoising, LV reconstruction, and CT-free AC in cardiac SPECT.
• The multi-task learning framework effectively addresses the challenges of improving cardiac SPECT imaging quality and reducing radiation dose.
• This approach has the potential to significantly advance the diagnosis of coronary artery diseases using SPECT imaging.