WDK-Net: Lightweight Wavelet Diffusion with Kolmogorov-Arnold Network for Limited-angle Cardiac CT Reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Medicine

Background:

• Limited-angle cardiac CT reconstruction suffers from artifacts and distortions due to incomplete angular data.
• Current diffusion-based methods offer high reconstruction quality but are computationally intensive, hindering clinical application.

Purpose of the Study:

• To develop a computationally feasible diffusion-based reconstruction method for limited-angle cardiac CT.
• To address the high computational cost and slow inference of existing diffusion models.

Main Methods:

• Proposed WDK-Net, a lightweight dual-domain reconstruction framework utilizing wavelet transforms.
• Implemented a structure-detail decoupled design to model global structures and local details in different frequency domains.
• Leveraged frequency-selective low-dimensional wavelet diffusion for efficient generative modeling.

Main Results:

• WDK-Net achieved competitive or superior reconstruction quality across various limited-angle settings (60°, 90°, 120°).
• Demonstrated substantial reduction in computational burden compared to existing diffusion-based methods.
• Validated performance on simulated, in-house clinical, and public cardiac CT datasets.

Conclusions:

• WDK-Net offers a computationally efficient and effective solution for limited-angle cardiac CT reconstruction.
• The proposed structure-detail decoupling and wavelet diffusion approach facilitates clinical deployment.
• This method presents a feasible pathway toward more accessible diffusion-based cardiac CT imaging.