Data-driven estimation of noise variance stabilization parameters for low-dose x-ray images

Area of Science:

• Medical Imaging
• Signal Processing
• Computational Science

Background:

• X-ray image denoising often uses noise statistics or noise variance stabilization (NVS).
• Generalized Anscombe transform (GAT) is an effective NVS technique, but requires system gain and electronic noise variance.
• These parameters are challenging to determine in clinical settings due to factors like beam hardening.

Purpose of the Study:

• To propose a data-driven method for estimating GAT parameters.
• To enable robust noise variance stabilization for improved X-ray image denoising.
• To address the challenge of predicting system gain and electronic noise in clinical X-ray imaging.

Main Methods:

• Utilized the energy compaction property of the discrete cosine transform.
• Employed a robust regression approach based on a linear Poisson-Gaussian model.
• Validated experimentally against beam hardening and denoising performance across various dose and scatter levels.

Main Results:

• Estimated system gain and electronic noise level with an average error of 4.2% in low-dose X-ray settings.
• Achieved performance gains of 5% in peak-signal-to-noise ratio and 4% in structural similarity index after GAT denoising.
• Demonstrated robustness across different dose and scatter levels.

Conclusions:

• Data-driven estimation of GAT parameters is efficient and robust.
• Improved parameter estimation leads to more accurate GAT-based NVS.
• This facilitates enhanced denoising of low-dose X-ray images using algorithms for additive Gaussian noise.