Deep learning-based reconstruction may improve non-contrast cerebral CT imaging compared to other current...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Iterative reconstruction algorithms are crucial for CT image quality.
• Deep learning reconstruction (DLR) is an emerging technique for CT image processing.
• Evaluating DLR against established methods like Hybrid-IR and MBIR is essential for clinical adoption.

Purpose of the Study:

• To compare the image quality and reconstruction times of a commercial DLR algorithm against Hybrid-IR and MBIR for cerebral non-contrast CT (NCCT).

Main Methods:

• Fifty cerebral NCCT scans were reconstructed using DLR, Hybrid-IR, and MBIR.
• Image quality was assessed subjectively by five observers and objectively via noise magnitude and SDNR.
• Reconstruction times for each method were recorded.

Main Results:

• DLR achieved significantly better subjective image quality scores than MBIR.
• DLR demonstrated lower noise (5.6) and higher SDNR (2.4) compared to Hybrid-IR (noise 6.4, SDNR 1.9) and MBIR (noise 6.2, SDNR 2.0).
• Reconstruction times were 27s (Hybrid-IR), 44s (DLR), and 176s (MBIR).

Conclusions:

• DLR provides superior image quality, with reduced noise and enhanced tissue differentiation compared to Hybrid-IR, at a slightly increased reconstruction time.
• DLR significantly outperforms MBIR in image quality, with substantially shorter reconstruction times.
• DLR represents a promising advancement for cerebral NCCT imaging.