DDT-Net: Dose-Agnostic Dual-Task Transfer Network for Simultaneous Low-Dose CT Denoising and Simulation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning (DL) shows promise for low-dose CT (LDCT) imaging.
• Existing DL denoising models struggle with generalizing to varied dose levels.
• LDCT simulation tools often require proprietary data, limiting accessibility.

Purpose of the Study:

• To develop a dose-agnostic network for simultaneous LDCT denoising and simulation.
• To address the generalization limitations of current DL denoising models.
• To create an accessible tool for simulating LDCT images at arbitrary dose levels.

Main Methods:

• Proposed DDT-Net, a dual-task transfer network integrating denoising and simulation.
• Employed a unified optimization framework to learn the joint distribution of LDCT and normal-dose CT (NDCT) data.
• Utilized a mixed-dose training strategy and approximated continuous dose levels with discrete data.

Main Results:

• DDT-Net demonstrated superior denoising and generalization performance on unseen dose data compared to existing methods.
• The simulated paired dataset effectively augmented LDCT images, restoring tissue texture.
• The network successfully simulated realistic LDCT images across arbitrary dose levels.

Conclusions:

• DDT-Net offers a robust solution for dose-agnostic LDCT denoising and simulation.
• The model overcomes generalization challenges and provides a valuable simulation tool.
• This approach enhances LDCT image quality and accessibility for research and clinical applications.