Micro-CT image denoising with an asymmetric perceptual convolutional network.

Area of Science:

• Biomedical Imaging
• Medical Physics
• Computer Vision

Background:

• Micro-computed tomography (Micro-CT) enables high-precision, non-invasive 3D imaging crucial for biomedical research.
• Acquiring high signal-to-noise ratio (SNR) images in Micro-CT is challenging under temporal resolution constraints due to limited radiation source power.
• Low-dose CT image denoising is essential for enhancing Micro-CT image quality while maintaining temporal resolution.

Purpose of the Study:

• To develop an advanced denoising method for low-dose Micro-CT images.
• To improve the capability of capturing and retaining fine image details in Micro-CT scans.
• To address the trade-off between image quality and acquisition time in Micro-CT.

Main Methods:

• An end-to-end asymmetric perceptual convolutional network (APCNet) was designed.
• The network architecture incorporates improvements to the convolutional layer for enhanced feature extraction.
• An edge detection layer was integrated to better preserve image details and structures.

Main Results:

• APCNet demonstrated superior performance in numerical metrics compared to established denoising models (DnCNN, CNN-VGG, RED-CNN).
• Visual perception assessments confirmed the enhanced image quality and detail preservation achieved by APCNet.
• The proposed method effectively reduces noise in low-dose Micro-CT images.

Conclusions:

• APCNet offers a significant advancement in low-dose Micro-CT image denoising.
• The network's design effectively balances noise reduction with the preservation of critical image micro-architecture.
• This method holds promise for improving the diagnostic utility of Micro-CT in biomedical applications.