Benchmarking deep learning-based low-dose CT image denoising algorithms.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Efforts to reduce radiation dose in computed tomography (CT) while maintaining image quality are ongoing.
• Iterative reconstruction and noise reduction algorithms are established techniques.

Purpose of the Study:

• To address the lack of standardized benchmarks and inconsistencies in experimental designs for deep learning-based CT denoising methods.
• To improve the verifiability and reproducibility of research in this field.

Main Methods:

• A standardized benchmark setup was developed for evaluating deep learning denoising algorithms.
• A comprehensive and fair evaluation of state-of-the-art methods was performed using the proposed setup.

Main Results:

• Most deep learning-based denoising methods demonstrated statistically similar performance.
• Improvements in performance over recent years were found to be marginal.

Conclusions:

• A need for more rigorous and fair evaluation of deep learning methods for low-dose CT image denoising is highlighted.
• The proposed benchmark setup serves as a foundational tool for future research and algorithm evaluation.