Impact of a deep learning reconstruction algorithm on image quality and dose reduction with ultra-high-resolution CT...

Area of Science:

• Medical Imaging
• Radiology
• Image Reconstruction

Background:

• Deep learning reconstruction (DLR) algorithms are emerging tools in medical imaging.
• Ultra-high-resolution (UHR) detectors offer enhanced spatial detail.
• Task-based image quality assessment is crucial for evaluating imaging systems.

Purpose of the Study:

• To quantitatively evaluate image quality and radiation dose reduction potential.
• To assess the combination of DLR algorithms with UHR detectors.
• To utilize a task-based assessment framework including MTF, NPS, TTF, and detectability index (d').

Main Methods:

• A Catphan 600 phantom was scanned at various CTDIvol levels.
• Data were reconstructed using Filtered Back-Projection (FBP), adaptive statistical iterative reconstruction (ClearView), and ClearInfinity (DLR).
• Modulation Transfer Function (MTF), Noise Power Spectrum (NPS), Task Transfer Function (TTF), and Detectability Index (d') were measured.

Main Results:

• UHR detectors (0.3125 mm collimation) improved spatial resolution (MTF) and detectability of small features but increased noise.
• DLR (ClearInfinity) reduced noise peaks without altering spatial frequency, enhancing detectability of large and subtle features.
• The combination of DLR and UHR detectors yielded the lowest noise and highest detectability index.

Conclusions:

• Integrating DLR with UHR detectors enhances spatial resolution and lesion detectability.
• This combination effectively reduces noise magnitude without texture alteration.
• Substantial potential for radiation dose reduction in medical imaging is demonstrated.