Evaluation of four computed tomography reconstruction algorithms using a coronary artery phantom

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Coronary computed tomography angiography (CTA) faces limitations in positive predictive value and specificity due to spatial resolution constraints.
• Advanced image reconstruction techniques are crucial for overcoming these challenges in coronary CTA.

Purpose of the Study:

• To evaluate the impact of second-generation deep learning-based reconstruction (DLR) on coronary CTA image quality.
• To compare the quantitative and qualitative performance of 2nd generation DLR against other reconstruction methods.

Main Methods:

• A vessel model with non-calcified plaque was scanned using 320-detector CT.
• Image reconstruction was performed using hybrid iterative reconstruction (HIR), model-based iterative reconstruction (MBIR), DLR, and 2nd generation DLR.
• Quantitative analysis included contrast-to-noise ratio (CNR) and edge rise slope (ERS); qualitative analysis involved observer grading of image properties.

Main Results:

• Second-generation DLR demonstrated significantly lower image noise (9.5 HU) compared to HIR, MBIR, and DLR.
• CNR and ERS values were highest with 2nd generation DLR (38.3 and 262.4 HU/mm, respectively).
• Subjective image quality scores for graininess, sharpness, and overall lumen visibility were superior with 2nd generation DLR.

Conclusions:

• Second-generation DLR significantly enhances CNR and ERS in coronary CTA.
• This advanced reconstruction method improves subjective image quality for assessing vessel stenosis.
• 2nd generation DLR offers superior performance over HIR, MBIR, and previous DLR techniques.