Low-dose CT denoising using a Progressive Wasserstein generative adversarial network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Low-dose computed tomography (LDCT) reduces patient radiation exposure but introduces image noise and artifacts.
• Image quality degradation significantly impacts clinical diagnosis, necessitating effective denoising techniques.
• Generative Adversarial Networks (GANs) show promise for LDCT denoising, but their complexity hinders analysis.

Purpose of the Study:

• To develop a more feasible and effective method for LDCT denoising.
• To address the complexity of existing GAN architectures for medical image processing.
• To improve the balance between artifact reduction and detail preservation in denoised LDCT images.

Main Methods:

• Proposed a progressive Wasserstein generative adversarial network (PWGAN) for LDCT denoising.
• Implemented recursive computation to reduce network parameters.
• Introduced a novel hybrid loss function, weighted structurally-sensitive hybrid loss function (WSHL), to minimize artifacts and retain image details.

Main Results:

• The proposed PWGAN-WSHL model demonstrated superior performance compared to state-of-the-art methods.
• Achieved significant reduction in image noise and artifacts.
• Successfully preserved crucial details in the denoised LDCT images.

Conclusions:

• The PWGAN-WSHL offers a simpler yet effective baseline for LDCT denoising.
• The novel hybrid loss function enhances artifact reduction while preserving image fidelity.
• This approach provides a promising solution for improving diagnostic accuracy in low-dose CT imaging.