AI-based Synthetic CT Angiography from Non-contrast CT for Cerebrovascular Disease Screening

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cerebrovascular Disease

Background:

• CT angiography (CTA) is crucial for cerebrovascular disease diagnosis but has limitations like availability and contrast agent use.
• Artificial intelligence (AI) offers potential to extract more information from standard imaging.
• This study explores generating CTA-like images from non-contrast CT (NCCT) using AI.

Purpose of the Study:

• To investigate the feasibility of deriving CT angiography (CTA) maps from non-contrast CT (NCCT) scans.
• To develop and validate an AI model for synthesizing CTA images from NCCT data.
• To assess the clinical utility of AI-generated CTA for cerebrovascular disease screening and triage.

Main Methods:

• A retrospective, multicenter study involving 3,709 patients with NCCT and CTA scans.
• Development of the Cerebrovascular CTA Generative Artificial Intelligence Model (CTA-GAI) to synthesize CTA from NCCT.
• Evaluation of synthetic CTA using quantitative metrics, visual assessment by clinicians, and clinical utility tests for disease classification.

Main Results:

• CTA-GAI consistently synthesized high-quality CTA-like images from NCCT, validated across multiple centers.
• Quantitative metrics (MAE, MSE, PSNR, SSIM) showed close approximation to real CTA images.
• Clinicians achieved high accuracy (92.7%) in distinguishing diseased from normal intracranial arteries and classifying subtypes (90.7% for IA vs. IAS).

Conclusions:

• AI-generated CTA from NCCT shows significant promise for preliminary cerebrovascular disease assessment.
• CTA-GAI can serve as a rapid, low-cost, non-invasive tool for large-scale screening and triage.
• This AI approach could enhance early detection and management of cerebrovascular conditions.