Accurate Image Domain Noise Insertion in CT Images

Area of Science:

• Medical Imaging
• Radiology
• Image Processing

Background:

• Simulating lower dose, noisy computed tomography (CT) images aids in optimizing protocols by assessing dose-image quality trade-offs.
• Existing noise insertion techniques often rely on proprietary projection data, limiting accessibility and scalability.

Purpose of the Study:

• To develop and validate an image domain approach for accurate CT noise insertion and low-dose scan simulation.
• To provide an accessible alternative to projection domain methods for noise simulation.

Main Methods:

• Utilizes image information to estimate variance maps and local noise power spectra (NPS).
• Applies filtered normally distributed noise in small image patches using inverse Fourier transform of the square root of local NPS.
• Generates spatially correlated noise by overlapping patches and scaling with a standard deviation map.

Main Results:

• Demonstrates excellent agreement with traditional projection domain methods for both simulated and real datasets.
• Successfully generates locally varying, spatially correlated noise.
• The method accurately scales noise based on dose relationships.

Conclusions:

• The proposed image domain framework offers an accurate and practical alternative for CT noise simulation.
• This method is particularly useful when projection domain techniques are not feasible, such as in large-scale studies.
• It enhances the toolkit for CT protocol optimization and image quality assessment.