Low-dose CT Denoising Using Edge Detection Layer and Perceptual Loss.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Low-dose computed tomography (CT) reduces patient radiation exposure but introduces image noise and artifacts.
• Deep neural networks (DNNs) show promise in denoising low-dose CT images.
• Existing DNNs may struggle with preserving fine details and precise boundaries.

Purpose of the Study:

• To enhance the performance of deep neural networks for low-dose CT image denoising.
• To introduce novel, computationally efficient techniques to improve image reconstruction quality.
• To address limitations of standard loss functions in CT image enhancement.

Main Methods:

• A non-trainable edge detection layer was integrated to extract edge maps, improving boundary definition.
• A joint objective function combining mean-square error (MSE) and perceptual loss was developed.
• The combined loss function mitigates over-smoothing from MSE and checkerboard artifacts from perceptual loss.

Main Results:

• The proposed edge detection layer improved quantitative metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM).
• The network successfully predicted CT images with more precise boundaries.
• The joint objective function effectively balanced noise reduction, detail preservation, and artifact suppression.

Conclusions:

• The novel techniques enhance low-dose CT image quality with minimal increase in network complexity.
• The approach offers a promising solution for improving diagnostic accuracy in low-dose CT imaging.
• This work contributes to safer and more effective medical imaging practices.