Impact of CT Acquisition Parameters on Deep Learning of Aortic Segmentation Performance: Systematic Review

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Automatic segmentation of computed tomography (CT) images is crucial for quantitative anatomical analysis in clinical settings.
• Artificial intelligence (AI) models show promise for aortic segmentation, but their performance is sensitive to CT image quality.

Purpose of the Study:

• To review how variations in CT acquisition parameters influence image quality and the accuracy of AI-based aortic segmentation models.
• To identify optimal CT acquisition strategies for reliable AI-driven aortic segmentation.

Main Methods:

• A narrative review following PRISMA guidelines was conducted.
• 13 studies were included, assessing the impact of CT technical parameters on image quality and AI-based aortic segmentation accuracy.

Main Results:

• Thinner CT slices (1 mm) significantly improved segmentation accuracy (Dice Similarity Coefficient [DSC] up to 0.87) and reduced errors.
• Faster CT acquisitions decreased image noise, and optimal pixel spacing enhanced image fidelity.
• 2D-3D U-Net ensembles achieved high accuracy (DSC = 0.928), though thrombus segmentation (DSC = 0.782) was more accurate than vessel segmentation (DSC = 0.481).

Conclusions:

• CT acquisition parameters critically influence the reliability of AI-driven aortic segmentation.
• Optimizing acquisition parameters can improve AI model performance, but challenges in AI interpretability necessitate standardized protocols and explainable AI.