Effect of a new deep learning image reconstruction algorithm for abdominal computed tomography imaging on image...

Area of Science:

• Medical imaging physics
• Radiology
• Artificial intelligence in healthcare

Background:

• Deep learning reconstruction algorithms aim to mitigate image texture changes from iterative methods.
• Evaluating Advanced intelligent Clear-IQ Engine (AiCE) against AIDR 3D and FIRST algorithms is crucial.

Purpose of the Study:

• To assess AiCE's impact on image quality and radiation dose.
• Compare AiCE with hybrid iterative reconstruction (AIDR 3D) and model-based iterative reconstruction (FIRST).

Main Methods:

• Utilized ACR 464 phantom across six dose levels (1-15 mGy).
• Reconstructed data using AIDR 3D, FIRST, and AiCE at varying strengths.
• Analyzed noise-power spectrum (NPS), task-based transfer function (TTF), and detectability index for simulated lesions.

Main Results:

• AiCE demonstrated lower NPS peaks and altered spatial frequencies compared to AIDR 3D and FIRST.
• Higher TTF values for AiCE on acrylic inserts, but varied performance on bone inserts.
• Improved detectability index for simulated lesions with AiCE, suggesting enhanced lesion visualization.
• Significant dose reduction potential observed, e.g., -79% for calcification and -57% for mass with AiCE at Standard level.

Conclusions:

• AiCE yields superior image quality with reduced noise and enhanced detectability over AIDR 3D and FIRST.
• Phantom study indicates significant dose reduction potential with AiCE.
• Clinical validation in patients is recommended to confirm findings.