Improved image quality with deep learning reconstruction - a study on a semi-anthropomorphic upper-abdomen phantom

Area of Science:

• Medical imaging
• Radiology
• Image processing

Background:

• Computed tomography (CT) reconstruction algorithms aim to balance image quality with radiation dose.
• Filtered back projection (FBP) is a basic method, while iterative reconstruction (IR) offers improvements but can introduce artifacts.
• Deep learning reconstruction (DLR) is an emerging technique with the potential to enhance image quality and reduce noise.

Purpose of the Study:

• To evaluate the image quality of a novel deep learning reconstruction (DLR) algorithm.
• To compare DLR performance against filtered back projection (FBP) and hybrid iterative reconstruction (IR) across various radiation dose levels.
• To assess the potential for radiation dose reduction using DLR in abdominal CT imaging.

Main Methods:

• A semi-anthropomorphic upper-abdominal phantom was scanned at five dose levels (5-25 mGy CTDIvol).
• Scans were reconstructed using FBP, hybrid IR (IR50, IR70, IR90), and DLR (low, medium, high strength) in 0.625 mm and 2.5 mm slices.
• Image quality metrics including CT number, noise, contrast, CNR, NPS, and TTF were analyzed.

Main Results:

• CT numbers were consistent across all reconstruction methods.
• DLR significantly reduced image noise, with higher DLR strength yielding greater noise reduction.
• Noise texture (NPS, NTD) was preserved with DLR, unlike hybrid IR, and DLR showed potential for 35-74% dose reduction compared to IR50.

Conclusions:

• The DLR algorithm effectively reduces noise while preserving image texture, surpassing limitations of traditional IR.
• DLR demonstrates significant potential for radiation dose reduction in abdominal CT.
• Thin-slice reconstruction with DLR may offer additional imaging benefits.