Image Recovery from Synthetic Noise Artifacts in CT Scans Using Modified U-Net

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-dose Computed Tomography (CT) scans are crucial for cancer screening but suffer from noise artifacts due to low photon counts.
• These noise artifacts reduce image clarity, potentially hindering accurate cancer detection by radiologists.
• Effective noise removal is essential to enhance the diagnostic performance of low-dose CT imaging.

Purpose of the Study:

• To develop and evaluate a novel Convolutional Neural Network (CNN) model for noise removal in low-dose CT images.
• To improve image quality and detail clarity compared to existing denoising techniques.
• To assess the model's performance across different noise densities and compare it with other CNN architectures and classical denoising methods.

Main Methods:

• A new CNN model, featuring a stacked modified U-Net architecture with optimized feature maps, was designed for image denoising.
• The model was trained and tested on a dataset of 174 low-dose CT images.
• Performance was evaluated using Peak Signal-to-Noise Ratio (PSNR) scores and visual assessment, with comparisons against other CNN models and classical denoising approaches.

Main Results:

• The proposed CNN model demonstrated superior denoising performance, achieving a higher average PSNR quality score compared to other CNN models.
• Visual analysis confirmed that the model's denoised images closely resemble full-dose CT scans, indicating significant detail restoration.
• The model proved effective in handling varying noise densities, as validated by testing on separate datasets.

Conclusions:

• The developed stacked modified U-Net CNN model offers an effective solution for noise removal in low-dose CT scans.
• This advancement can enhance image clarity and potentially improve cancer detection rates in medical imaging.
• The proposed method shows promise for clinical application, outperforming traditional denoising techniques in complex scenarios.