Lesion Insertion in Projection Domain for Computed Tomography Image Quality Assessment

Area of Science:

• Medical Imaging
• Radiology
• Computational Imaging

Background:

• Task-based image quality assessment in CT requires realistic patient images with known diagnostic truth.
• Conventional image-domain hybrid image generation does not accurately represent the impact of scan and reconstruction parameters on simulated lesions.

Purpose of the Study:

• To develop and validate a projection-domain approach for generating realistic hybrid CT images with simulated liver lesions.
• To improve the accuracy of lesion appearance simulation for task-based CT image quality assessment.

Main Methods:

• Liver lesion models were forward projected using CT scanner geometry to obtain lesion projections.
• Lesion projections were inserted into patient projections (from raw CT data) and reconstructed to create hybrid images.
• Validation involved comparing simulated ACR phantom images with physically acquired ones and visually assessing hybrid images.

Main Results:

• Simulated and physically acquired phantom images showed high similarity in Hounsfield Unit (HU) accuracy and high-contrast resolution.
• Hybrid images featured realistic lesions that integrated naturally into the liver background.
• Simulated lesions exhibited appearance variations dependent on reconstruction parameters.

Conclusions:

• The projection-domain method generates more realistic hybrid CT images than conventional image-domain techniques.
• This advanced method is suitable for creating datasets for task-based CT image quality assessment.
• The approach accurately models the influence of reconstruction parameters on lesion appearance.