ActiveNaf: A novel NeRF-based approach for low-dose CT image reconstruction through active learning

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Healthcare

Background:

• Conventional CT imaging poses radiation risks for patients needing repeat scans.
• There is a critical need for effective dose-reduction techniques in CT imaging.
• Maintaining image quality during dose reduction is a significant challenge.

Purpose of the Study:

• To develop a method for reducing radiation doses in CT imaging without compromising image quality.
• To optimize CT reconstructions using a limited number of X-ray projections.
• To combine Neural Attenuation Fields (NAF) with active learning for improved CT reconstruction.

Main Methods:

• A secondary neural network predicts the Peak Signal-to-Noise Ratio (PSNR) of 2D projections generated by NAF.
• An active learning strategy utilizes PSNR predictions to select the most informative X-ray projections.
• An iterative projection acquisition process is employed, contrasting with conventional single-session acquisition.

Main Results:

• The proposed method achieves high-quality 3D CT reconstructions from sparse data.
• Significant improvements in image quality metrics (PSNR3D, SSIM3D, PSNR2D) were observed compared to the baseline.
• The method attained equivalent image quality using 36 projections compared to the baseline's 60 projections.

Conclusions:

• The approach enables high-quality 3D CT reconstructions with significantly reduced radiation exposure.
• Clearer and more detailed anatomical images are produced from sparse projection data.
• This work advances safer and more efficient medical imaging procedures.