DaNet: dose-aware network embedded with dose-level estimation for low-dose CT imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Deep learning (DL) methods for low-dose CT (LDCT) image restoration often ignore radiation dose variations.
• Radiation dose differences significantly impact restoration quality, with lower doses posing greater challenges.
• Clinical practice requires methods that can adapt to varying scanning doses and estimate acceptable levels.

Purpose of the Study:

• To propose an adaptive dose-aware network for LDCT image restoration that accounts for input dose differences.
• To develop a two-stage approach for improved image quality and diagnostic accuracy in LDCT.

Main Methods:

• A novel two-stage deep learning network is proposed, starting with dose level estimation.
• Five distinct radiation dose levels (lowest to highest) were defined for simulation.
• The estimated dose level guides image restoration via channel feature transform in the second stage.

Main Results:

• Experiments on simulated data show the proposed method outperforms existing DL-based techniques.
• Ablation studies confirm the effectiveness of the dose-level estimation component.
• The method achieves superior peak signal-to-noise ratio and visual quality.

Conclusions:

• The adaptive dose-aware network effectively addresses dose variations in LDCT image restoration.
• The proposed method offers improved image quality and diagnostic potential for LDCT.
• Future work will focus on clinical validation across different CT equipment vendors.