CCN-CL: A content-noise complementary network with contrastive learning for low-dose computed tomography denoising.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Low-dose computed tomography (LDCT) is crucial for reducing radiation exposure in medical imaging.
• Current deep learning denoising methods for LDCT primarily rely on normal-dose CT (NDCT) images as positive training examples.
• Contrastive learning principles suggest that incorporating original low-dose images as negative examples can enhance network training.

Purpose of the Study:

• To propose a novel content-noise complementary network with contrastive learning (CCN-CL) for enhanced LDCT denoising.
• To leverage both NDCT images (positive) and original LDCT images (negative) within a contrastive learning framework.
• To improve the performance of deep learning-based LDCT denoising.

Main Methods:

• Developed a novel content-noise complementary network incorporating contrastive learning.
• Implemented a contrastive learning loss function using NDCT images as positive and LDCT images as negative examples.
• Integrated an attention mechanism and deformable convolution within the network architecture.

Main Results:

• The proposed CCN-CL network demonstrated superior performance compared to state-of-the-art methods on the 2016 NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge dataset.
• Quantitative and qualitative evaluations confirmed the effectiveness of the developed denoising method.
• The network successfully reduced noise while preserving diagnostic image quality in LDCT scans.

Conclusions:

• The novel CCN-CL network is a feasible and effective deep learning-based method for LDCT denoising.
• The integration of contrastive learning and a specialized network architecture significantly improves denoising performance.
• This approach offers a promising solution for enhancing the diagnostic utility of LDCT imaging.