Low-dose CT Denoising with Dilated Residual Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-dose Computed Tomography (CT) reduces radiation risk but degrades image quality.
• Noise removal techniques are crucial for improving low-dose CT image fidelity.
• Convolutional neural networks show potential for denoising CT images.

Purpose of the Study:

• To propose a deep residual network with dilated convolution for enhanced low-dose CT image denoising.
• To improve the quality of reconstructed images from low-dose CT scans.
• To reduce computational costs and layers while maintaining performance.

Main Methods:

• A deep residual network architecture incorporating identity mappings for signal propagation.
• Utilization of dilated convolutions to rapidly expand the receptive field.
• End-to-end learning from low-dose to normal-dose CT image mapping.

Main Results:

• The proposed network effectively denoises low-dose CT images, enhancing image quality.
• Identity mappings improved network performance and reduced training time.
• Dilated convolutions enabled achieving good results with fewer layers and lower computational expense.

Conclusions:

• The developed deep residual network with dilated convolution is effective for low-dose CT image denoising.
• This approach offers a promising solution for improving diagnostic accuracy in CT scans with reduced radiation.
• The method balances image quality enhancement with computational efficiency.