Computational and human observer image quality evaluation of low dose, knowledge-based CT iterative reconstruction

Area of Science:

• Medical Imaging Physics
• Computational Observer Models
• Radiological Dose Optimization

Background:

• Filtered back projection (FBP) is a standard computed tomography (CT) reconstruction method.
• Knowledge-based iterative reconstruction (IMR) offers potential improvements in image quality and radiation dose reduction.
• Accurate prediction of human observer performance is crucial for evaluating new imaging techniques.

Purpose of the Study:

• To develop a computational model observer for tracking human detectability in low-dose CT images reconstructed with IMR and FBP.
• To utilize the model for evaluating detectability trends and predicting human observer performance across various imaging parameters.
• To validate the model's predictions through human observer studies and demonstrate its application in CT imaging optimization.

Main Methods:

• A five-channel channelized Hotelling observer (CHO) model was developed, incorporating internal noise models.
• Model parameters were optimized using maximum likelihood estimation on phantom study data across varying dose, size, and contrast.
• Semianalytic internal noise computation was employed to accelerate parameter estimation and model validation.

Main Results:

• The developed CHO model (Model-k4) accurately predicted human observer detectability in blended FBP-IMR images.
• Iterative reconstruction (IMR) demonstrated an average detectability improvement of 2.7 times over FBP across all conditions.
• The model predicted an 82% dose reduction potential with IMR, subsequently verified by 80% dose reduction in physical CT scans.

Conclusions:

• Iterative reconstruction (IMR) significantly enhances detectability in low-dose CT compared to FBP, enabling substantial radiation dose reduction.
• A channelized Hotelling observer model with specific internal noise characteristics effectively mimics human observer performance across diverse imaging conditions.
• The validated model observer is a valuable tool for optimizing iterative reconstruction algorithms and guiding dose reduction protocols in clinical CT practice.