Clinical validation of an AI-based motion correction reconstruction algorithm in cerebral CT

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Cerebral CT Analysis

Background:

• Motion artifacts are a common problem in cerebral CT scans, often leading to reduced image quality and diagnostic uncertainty.
• Rescans due to motion artifacts increase patient radiation exposure and healthcare costs.
• Existing reconstruction algorithms may struggle to effectively mitigate motion-induced degradation.

Purpose of the Study:

• To evaluate the clinical performance of an AI-based motion correction (MC) reconstruction algorithm for cerebral CT.
• To quantitatively and qualitatively assess the algorithm's ability to reduce motion artifacts.
• To determine if the AI-MC algorithm improves diagnostic accuracy in cerebral CT.

Main Methods:

• Retrospective analysis of 53 cerebral CT cases with initial motion artifacts requiring rescans.
• Comparison of three image reconstruction groups: hybrid iterative reconstruction (IR) for the initial scan (motion group), hybrid IR for the rescan (reference group), and AI-based MC for the initial scan (MC group).
• Evaluation of image quality using objective metrics (SNR, CNR, MSE, PSNR, SSIM, MI) and subjective assessments, alongside diagnostic performance metrics like lesion detectability and Alberta Stroke Program Early CT Score (ASPECTS).

Main Results:

• The AI-MC group showed significantly increased SNR and CNR compared to the motion group.
• Objective image quality metrics (MSE, PSNR, SSIM, MI) were significantly improved in the MC group versus the reference group (44.1%, 15.8%, 7.4%, 18.3% improvement, respectively; p < 0.001).
• Subjective image quality scores were higher for the MC group, and diagnostic performance improved with higher lesion detectability and AUC in ASPECTS assessment (0.817 vs 0.614) compared to the motion group.

Conclusions:

• The AI-based MC reconstruction algorithm is clinically validated for reducing motion artifacts in cerebral CT.
• The algorithm demonstrably improves both quantitative and qualitative image quality and enhances diagnostic confidence for brain lesions.
• Implementing this AI-MC algorithm can potentially reduce the need for rescans, thereby improving the efficiency of emergency department cerebral CT workflows.