Deep learning-based image reconstruction for brain CT: improved image quality compared with adaptive statistical...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Computed tomography (CT) is crucial for brain imaging.
• Image reconstruction techniques aim to improve diagnostic quality while minimizing radiation dose.
• Adaptive statistical iterative reconstruction-Veo (ASIR-V) is a current standard, but deep learning-based image reconstruction (DLIR) offers potential advancements.

Purpose of the Study:

• To compare the image quality of brain CT scans reconstructed using DLIR versus ASIR-V.
• To evaluate objective and subjective image quality parameters between the two reconstruction methods.

Main Methods:

• Sixty-two patients underwent noncontrast brain CT scans.
• Datasets were reconstructed using 30% ASIR-V and DLIR at low, medium, and high strength levels.
• Objective parameters (CT attenuation, noise, artifact index, CNR) and subjective parameters (differentiation, sharpness, diagnostic quality) were assessed.

Main Results:

• DLIR demonstrated a significant, dose-dependent reduction in image noise and posterior cranial fossa artifacts compared to ASIR-V (P < 0.001).
• Contrast-to-noise ratio (CNR) improved with increasing DLIR strength levels in both centrum semiovale and basal ganglia (P < 0.001).
• DLIR with medium and high strengths yielded superior subjective image quality scores, with moderate to good interobserver agreement.

Conclusions:

• Optimized DLIR protocols significantly reduce noise and artifacts in routine brain CT scans.
• DLIR offers improved subjective image quality compared to ASIR-V for brain imaging.
• DLIR represents a promising advancement in CT image reconstruction for enhanced diagnostic accuracy.