CT iterative vs deep learning reconstruction: comparison of noise and sharpness

Area of Science:

• Radiology
• Medical Imaging
• Computational Imaging

Background:

• Iterative reconstruction techniques like adaptive statistical iterative reconstruction-V (ASIR-V) are standard for CT angiography.
• Deep learning reconstruction (DLR) methods, such as TrueFidelity (TFI), are emerging as advanced alternatives.
• Optimizing image quality in lower extremity CT angiography is crucial for accurate diagnosis and treatment planning.

Purpose of the Study:

• To compare image noise and sharpness in lower extremity CT angiography between ASIR-V and TFI.
• To evaluate the performance of different ASIR-V blending factors and TFI strength levels.
• To determine the optimal reconstruction method for balancing image noise and sharpness.

Main Methods:

• Thirty-seven patients underwent lower extremity CT angiography.
• Images were reconstructed using ASIR-V (80% and 100% blending) and TFI (low, medium, high strength).
• Quantitative (CT number, noise, SNR, CNR, blur metrics) and qualitative assessments by radiologists were performed on vessels, liver, and muscle.

Main Results:

• Increasing ASIR-V blending factors and TFI strength reduced image noise and increased SNR/CNR.
• TFI images demonstrated significantly lower blur metric values (higher sharpness) than ASIR-V images (p < 0.001).
• High-strength TFI (TF-H) yielded the highest SNR/CNR, comparable to 100% ASIR-V, with superior sharpness.

Conclusions:

• Deep learning reconstruction (TrueFidelity) is superior to iterative reconstruction (ASIR-V) for image noise and sharpness in lower extremity CT angiography.
• The high-strength TFI setting (TF-H) provided the most balanced image quality, optimizing both noise reduction and sharpness.
• Higher blending/strength factors in both ASIR-V and TFI generally lead to lower noise but reduced sharpness.