MLAR-UNet: LDCT image denoising based on U-Net with multiple lightweight attention-based modules and residual...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-dose CT (LDCT) is essential for reducing radiation exposure during medical imaging.
• However, LDCT images suffer from increased noise and artifacts, hindering accurate diagnosis.
• Existing denoising methods often struggle to balance noise reduction with detail preservation.

Purpose of the Study:

• To develop an advanced deep learning model for effective LDCT image denoising.
• To improve the preservation of crucial diagnostic details in low-dose CT scans.
• To introduce novel attention-based modules for enhanced image processing.

Main Methods:

• A U-Net based deep learning architecture, MLAR-UNet, was proposed.
• Integration of multiple lightweight attention modules: CBAM, CR, ACRM, and CTCAM.
• Novel modules ACRM and CTCAM were designed to enhance feature representation and attention mechanisms using Transformer and convolution.

Main Results:

• MLAR-UNet demonstrated superior performance in denoising LDCT images compared to state-of-the-art methods.
• The model effectively preserved image details and reduced noise in clinical chest and abdominal CT datasets.
• Experimental validation confirmed the efficacy and necessity of each integrated module.

Conclusions:

• The proposed MLAR-UNet offers a significant advancement in LDCT image denoising.
• The novel ACRM and CTCAM modules provide strong detail comprehension with minimal computational overhead.
• This work presents an efficient approach for integrating Transformer-based attention mechanisms in medical image processing.